South Korea’s Central Bank Weighs Next Moves as AI Boom Fuels Growth and Price Pressures

SEOUL — The Bank of Korea raised its benchmark interest rate twice in quick succession this summer. It now sits at 3%. Yet the debate over further tightening has only grown sharper.

Inflation refuses to settle near the 2% target. Growth has surprised to the upside, powered by semiconductor exports tied to the global surge in artificial intelligence. Household debt climbs. Housing prices in the Seoul area accelerate. Policymakers tread carefully.

On Tuesday, board member Chang Yong-sung laid out the framework. The central bank will set the timing and pace of additional rate hikes by watching inflation, economic growth and financial stability conditions. “Going forward, factors such as accumulating financial imbalances, sector-specific income improvements, global market trends and geopolitical risks will likely shape financial stability,” he said in remarks released with the Reuters report.

Chang stressed coordination. Monetary policy and macroprudential tools must work together. Fiscal and financial authorities need to address difficulties for vulnerable groups. The message was clear. No predetermined path exists. Data will decide.

BOK Shifts Stance After Years of Accommodation

The shift began in July. The Monetary Policy Board lifted the base rate 25 basis points to 2.75%. All seven members backed the move. It marked the first hike in three and a half years. Inflation had climbed above 3%. Semiconductor exports roared ahead. Domestic demand showed signs of recovery.

August brought another 25 basis point increase. The rate reached 3%. Six members voted yes. One dissented. The central bank revised its 2026 growth forecast sharply higher to 3.3% from 2.6%. The 2027 projection rose to 2.9%. Inflation forecasts held at 2.7% this year and 2.3% next. Core inflation, however, was marked up. It is now seen at 2.5% for both years.

Governor Shin Hyun Song explained the back-to-back action. Preemptive steps help anchor expectations. They limit the eventual cost of reining in prices. Yet he also signaled caution. The board would assess the effects of the two hikes before deciding on more.

By September, the tone remained measured. The BOK’s own Monetary Policy Report stated it would decide additional hikes while closely monitoring inflation, economic developments and financial stability. Inflation is projected to stay above target for a prolonged period. Growth should remain solid, supported by exports, investment and recovering consumption. Housing prices and household loans are both picking up speed. (Bank of Korea)

Short sentences. Clear risks. Persistent price pressures. Stronger domestic demand. These forces collide.

Analysts have taken notice. JPMorgan Chase stands out for its hawkish view. The bank sees upside to its already bold call for the policy rate to reach 3.75%. It expects hikes in November, February and May of next year. Semiconductor-driven expansion could stoke even stronger inflation, provided credit and financial markets hold steady. That terminal rate exceeds the 3.5% median in Bloomberg surveys. (Bloomberg, published Sept. 21, 2026)

But not everyone agrees on the pace. Minutes from the August meeting, released in mid-September, revealed divisions. Dissenter Hwang Kun-il argued for holding rates. He pointed to rising delinquencies and the need to support growth. A stronger won had given room to evaluate prior moves. Such splits suggest future decisions may come more slowly. (Reuters)

And markets have priced in more. Some forecasts see the 1-year forward 3-month rate near 4.2%. That implies roughly four hikes from current levels. Economists at Goldman Sachs see only one additional 25 basis point move, taking the rate to 3.25%. The gap between market pricing and bank projections creates trading opportunities in rates markets.

Geopolitics adds uncertainty. Tensions in the Middle East have kept oil prices elevated. That feeds imported inflation in an energy-dependent economy. The won’s path matters too. A weaker currency amplifies cost pressures. Recent strength helped, yet volatility persists.

Financial stability concerns loom large. Household debt in South Korea ranks among the highest relative to income in major economies. Apartment prices in the capital region have accelerated. Consumer loans grow. The central bank has warned repeatedly about these imbalances. Chang’s comments Tuesday reinforced the need for complementary policies to prevent them from deepening.

So the board watches incoming numbers. August and September inflation readings. Business sentiment. Nominal GDP. Spillovers from the semiconductor sector into wages and consumption. Each data point shapes the next vote.

The October meeting will test the mood. Many expect no change then. But the door stays open for November. One more hike this year would bring the rate to 3.25%. Further moves in 2027 remain possible if growth stays hot and prices fail to moderate.

Korea’s economy shows a clear split. Export powerhouses in chips thrive on AI demand. Domestic sectors lag. This bifurcation complicates policy. Strong headline growth gives room to tighten. Yet uneven gains risk leaving parts of the economy behind.

Shin has emphasized flexibility. Each decision stays live. No mechanical path forward. That approach fits the moment. Global central banks navigate similar crosscurrents. The Federal Reserve recently hiked. Other major banks adjust. Korea cannot ignore external forces.

Yet its choices remain homegrown. Assess the data. Balance growth against prices. Guard financial stability. Chang’s remarks Tuesday captured the essence. The central bank holds the tools. It will use them as conditions warrant. No more. No less.

Investors, businesses and households wait for the next signal. The semiconductor cycle could extend. Oil prices could spike again. Domestic demand could accelerate faster than expected. Any of these would tilt the board toward tighter policy.

For now, the stance is restrictive. Rates at 3% after two quick moves. Inflation forecasts above target. Growth upgraded. The BOK has moved from accommodation to restraint. How far it goes depends on the balance of those three factors — inflation, growth, stability. The coming months will reveal whether more hikes lie ahead or if the current level suffices to cool pressures.

from WebProNews https://ift.tt/iq1uKY8

Linux 7.3-rc4 Released with Fixes for x86, ARM64, Btrfs, EXT4 and More

The Linux kernel development cycle continues its steady pace with the release of version 7.3-rc4. Linus Torvalds announced the fourth release candidate on the kernel mailing list, noting that the update arrives with a typical mix of fixes across various subsystems while maintaining overall stability for what has become a routine part of the merge window process.

This latest candidate brings together contributions from hundreds of developers, addressing issues that surfaced after the initial code merges for the 7.3 series. According to the Phoronix report, the patch volume sits at a moderate level compared to previous candidates, suggesting that the kernel team has already resolved many of the larger structural changes and now focuses on polishing specific areas that showed problems during testing.

Several architecture-specific updates appear in this release. The x86 maintainers included corrections for memory management and virtualization components, particularly around Intel and AMD hardware support. These changes help ensure consistent behavior across different processor generations, especially when dealing with newer instruction set extensions and power management features. ARM64 developers contributed fixes related to interrupt handling and memory barrier semantics that could have caused rare but problematic race conditions on multi-core systems.

Storage and file system improvements form another significant portion of the changes. The Btrfs file system received multiple patches that address edge cases in its RAID implementation and snapshot handling. Users who rely on Btrfs for complex storage setups will benefit from these refinements, which reduce the chance of metadata corruption during heavy write workloads. EXT4 also saw attention, with updates that improve handling of inline data and directory indexing under high contention.

The block layer and SCSI subsystem picked up several important corrections. These address potential deadlocks and error handling paths that became apparent during stress testing with modern solid-state drives and high-performance storage arrays. The device mapper component, which underpins technologies like LVM and software RAID, received fixes that enhance reliability when resizing logical volumes or handling transient device failures.

Networking remains a constant area of activity in kernel development, and 7.3-rc4 continues that tradition. The TCP stack saw improvements to congestion control algorithms and packet scheduling, helping to reduce latency in certain network conditions. WiFi drivers for popular chipsets from Intel, MediaTek, and Realtek gained bug fixes that resolve connectivity drops and power management issues reported by users on laptops and embedded devices. The Bluetooth subsystem also received attention, particularly around Low Energy device pairing and audio streaming stability.

Graphics and display support continue to evolve with this candidate. The AMD graphics drivers incorporated corrections for power management transitions and display output handling on newer GPU generations. Intel’s i915 driver team addressed several race conditions that could lead to screen corruption or system instability when switching between different display configurations. The Nouveau driver for NVIDIA hardware picked up minor fixes that improve compatibility with recent Mesa userspace components.

One area that has seen consistent attention throughout the 7.3 cycle involves Rust integration within the kernel. While still considered experimental, the infrastructure for writing kernel modules in Rust has received refinements that make the binding layer more reliable. These changes primarily affect how Rust code interacts with core kernel APIs for memory allocation and synchronization primitives. The Phoronix coverage highlights that these updates represent incremental progress rather than major new capabilities, which aligns with the cautious approach the kernel community has taken toward introducing new programming languages.

Hardware support continues to expand in expected ways. The sound subsystem gained better compatibility with newer audio codecs found in recent laptops and mobile devices. USB updates address enumeration problems with certain high-speed peripherals, while the PCI subsystem received patches that improve resource allocation on systems with complex bridge configurations.

Performance monitoring and debugging tools also benefited from this release candidate. The perf subsystem incorporated fixes that restore accurate event counting on certain CPU models, which should please developers who rely on precise measurements for optimization work. The tracing infrastructure saw improvements to probe handling that reduce overhead when using kprobes on hot code paths.

Security-related changes appear throughout the candidate, though many take the form of hardening rather than fixes for newly discovered vulnerabilities. The memory management code received additional checks that help prevent certain classes of use-after-free errors. The SELinux and AppArmor security modules both saw refinements to their policy handling that eliminate potential inconsistencies when dealing with complex namespace setups.

Testing and quality assurance efforts have clearly influenced the content of this release candidate. The kernel test robot and various automated validation systems flagged issues that developers then addressed before this snapshot. This process helps catch problems that might otherwise only surface after wider deployment. The moderate size of the diffstat for rc4 suggests that the development cycle has moved past the period of major feature integration and into a phase where stability takes priority.

For those following the kernel release schedule, this candidate arrives roughly on the expected timeline. The merge window for 7.3 closed several weeks ago, and the community has since focused on integrating bug reports from early testers and continuous integration systems. If the pattern from previous cycles holds, we can expect one or two more release candidates before the final 7.3 version arrives, assuming no major regressions appear in the coming days.

The involvement of major technology companies remains evident in the contributor list. Engineers from Intel, AMD, Google, Red Hat, SUSE, and various hardware vendors have all submitted patches that made their way into this candidate. This broad participation reflects the diverse requirements that the Linux kernel must satisfy, from enterprise servers and cloud infrastructure to consumer laptops and mobile devices.

Users who track kernel development through git trees can pull this candidate immediately for testing. The tag linux-7.3-rc4 is available in the main kernel repository, and distribution maintainers will likely begin incorporating it into their development branches soon. Those running earlier candidates or the 7.2 stable series may want to consider upgrading if they have encountered any of the specific issues addressed in this update.

The ongoing development process demonstrates how the kernel community balances the introduction of new features with the need for reliability. Each release candidate serves as both a snapshot of current progress and an opportunity for wider testing before finalization. The 7.3 series appears to be following this established pattern, with rc4 providing another step toward what should be a solid release when it eventually arrives.

Looking at the broader context of kernel development, the 7.3 cycle has incorporated several notable changes beyond the bug fixes highlighted in this candidate. Earlier merge windows brought updates to scheduling algorithms, memory management improvements, and expanded support for newer hardware platforms. The Rust integration work, while still in its early stages, continues to expand the options available to developers who want to write safer kernel code without sacrificing performance.

The file system updates deserve particular attention because they affect data integrity across countless deployments. The improvements to Btrfs and EXT4 in this candidate build upon years of refinement that have made these file systems suitable for demanding production environments. Similar attention to the networking stack ensures that Linux remains competitive in high-throughput data center applications while maintaining the responsiveness needed for desktop and mobile use cases.

Driver updates for modern hardware continue to arrive at a steady pace. As new processors, graphics cards, and peripheral devices reach the market, kernel developers work to ensure compatibility and optimal performance. This rc4 release includes work that helps prepare the kernel for hardware that vendors have only recently announced, demonstrating the proactive nature of much kernel development work.

The security enhancements, though sometimes less visible to end users, play a vital role in maintaining trust in the platform. By continuously improving memory safety, access controls, and cryptographic primitives, the kernel team helps protect systems against both known and emerging threats. These changes often result from collaboration between academic researchers, industry security teams, and independent developers who identify potential weaknesses before they can be exploited.

Performance work in this candidate focuses on reducing latency and improving efficiency rather than introducing dramatic new optimizations. The accumulated effect of many small improvements across subsystems often leads to measurable gains when running real-world workloads. Users who test this candidate on their specific hardware configurations may notice smoother operation in areas that previously showed occasional hiccups.

The documentation updates included in rc4, while not always the most exciting changes, help ensure that both new and experienced kernel developers can understand the current state of the code. Updated comments, improved explanations in the documentation directory, and clearer error messages all contribute to a more maintainable codebase over time.

As the kernel approaches what may be its final release candidates, the focus will increasingly shift toward identifying and resolving any remaining issues that could affect stability. The community relies on users who run these pre-release versions to report problems promptly so developers can address them before the version number increments to the final release.

This steady progression through the release cycle reflects years of experience in managing a project of this scale and complexity. The Linux kernel has grown to support an enormous range of hardware and use cases, yet the development process remains structured around these regular merge windows and release candidates. Each step brings the project closer to delivering another version that will serve as the foundation for operating systems and embedded devices worldwide for months or even years to come.

The contributions that make up 7.3-rc4 represent countless hours of development, testing, and review by individuals and teams spread across different time zones and organizations. Their collective effort ensures that Linux continues to meet the diverse needs of its users while preparing for the technological demands of the future. As testing of this candidate expands, feedback from the broader community will help shape the final form of the 7.3 release and influence the direction of subsequent development cycles.

from WebProNews https://ift.tt/IcX7RBH

Microsoft: AI Licenses Alone Fail—Workflow Redesign Drives 20% Sales Boost and 75% Faster Supply Chains

Microsoft has spent the last several years treating its own workforce as a living laboratory for artificial intelligence adoption. The results, shared in a September 17 blog post by Chief People Officer Kathleen Hogan, show that simply handing out licenses produces limited returns. Real performance gains arrive only after teams redesign the underlying processes that define how work actually happens.

Hogan’s account stands out because it comes from inside the company that sells Copilot to the world. Rather than promotional language, the post offers a candid assessment of what happened when Microsoft turned the same tools on itself. The numbers cited are concrete: sales teams that rebuilt their deal workflows around AI recorded 20 percent higher close rates, supply chain groups shortened cycle times by as much as 75 percent, and one nine-person product team delivered a major release in just 35 days. These outcomes did not appear on adoption dashboards that tracked login counts or prompt volume. They surfaced only after leaders stopped measuring usage and started measuring changed business results.

The distinction between adoption and transformation sits at the center of the lessons Microsoft learned. Early in the rollout, the company distributed licenses to more than 100,000 employees. Usage climbed, yet many managers reported that daily rhythms felt unchanged. Meetings still ran long, decision loops remained slow, and output quality showed only marginal improvement. The pattern repeated across departments. Engineers used Copilot to accelerate code completion, but the broader development lifecycle stayed anchored to the same handoffs and approval gates that existed before the tools arrived. Sales representatives generated better first drafts of proposals, yet the end-to-end sales motion continued to follow legacy stages that no longer matched the speed of the new material.

That realization prompted a shift in focus. Instead of asking “how many people are using the tools,” Microsoft began asking “which workflows produce measurably better outcomes when AI is embedded inside them.” The question forced teams to map every step of their current process, identify which steps consumed the most time or introduced the most error, and then redesign those steps around AI capabilities. The resulting changes looked different in every function, but they shared a common trait: AI stopped being an optional assistant and became part of the standard operating procedure.

Nowhere was this redesign more visible than inside the sales organization. Traditional deal reviews involved lengthy slide decks, multiple alignment meetings, and manual updates to customer relationship management records. After the transformation, teams built prompts that automatically pulled customer history, competitor data, and pricing guardrails into a single living document. The document updated in real time as new information arrived, eliminating the need for separate status reports. Sales managers could query the document directly instead of scheduling yet another sync call. The compressed cycle allowed representatives to spend more time with customers and less time on internal coordination. The 20 percent lift in close rates emerged from that time reallocation, not from any single prompt trick.

Supply chain teams followed a similar path. Legacy processes relied on batch reporting that arrived days or weeks after events occurred. When disruptions hit, teams spent hours reconciling data from separate systems before they could even begin to model alternatives. The redesigned workflow connected Copilot directly to live telemetry streams from factories, ports, and logistics partners. Planners could ask natural-language questions about inventory buffers, lead-time variances, or alternative sourcing options and receive answers grounded in current data. The system flagged anomalies before they became crises. Cycle times that once stretched into multiple weeks shrank dramatically. In one documented case, a procurement team reduced the end-to-end sourcing process from 28 days to seven. The 75 percent reduction cited by Hogan reflects the aggregate impact across dozens of such improvements.

Product development offered perhaps the clearest illustration of speed gains. A small team tasked with modernizing an internal administrative portal decided to treat the entire build as an AI-augmented effort. Rather than writing detailed specifications and handing them to developers, the group maintained a living requirements document that Copilot referenced continuously. Developers described features in plain language, received working code snippets, and iterated inside the same chat thread. Quality assurance, normally a separate phase, ran in parallel because the model could generate test cases and edge-condition scenarios on demand. The result was a production-ready service delivered in 35 days by nine people, a timeline that would have been unthinkable under the previous staged approach.

These examples share more than impressive metrics. Each required leaders to relinquish the comfort of measuring activity and instead embrace accountability for outcomes. Adoption dashboards still exist inside Microsoft, but they now serve as diagnostic tools rather than success indicators. When usage in a particular team remains high yet business metrics stay flat, the diagnosis almost always points to an unchanged workflow. The remedy is not more training on prompt engineering. It is a structured redesign exercise that brings together the people who do the work, the leaders who own the results, and the AI specialists who understand the models’ current limits.

Hogan emphasizes that this redesign work is harder than distributing software. It demands psychological safety so employees will admit which parts of their current process are inefficient. It requires time from senior leaders who must participate in mapping sessions rather than simply endorsing them. And it forces uncomfortable trade-offs when legacy controls, designed to reduce risk, now slow down the very outcomes leaders claim to want. Many teams discovered that their most deeply held assumptions about necessary approvals or mandatory documentation crumbled under scrutiny once real-time AI verification became available.

The cultural dimension of this shift receives equal attention in the blog post. Microsoft had to retrain managers to evaluate performance based on delivered value rather than hours logged or documents produced. Some employees initially worried that visible AI assistance would make their contributions look smaller. Leaders countered by celebrating outcomes publicly and by sharing specific examples of how human judgment, not machine output, remained the decisive factor. A sales representative might use AI to generate a compelling proposal, but only the representative’s relationship insight could close the deal. A supply chain analyst might receive instant scenario models, but only the analyst’s experience with supplier behavior could select the right path. Making those distinctions explicit helped reduce anxiety and increased willingness to experiment.

Another lesson involved the pace of capability improvement. Microsoft’s own models advanced rapidly during the period covered by the blog post. Features that required custom development in the first quarter became native capabilities by the third. Teams that had invested weeks building bespoke connectors found those connectors suddenly unnecessary. The realization pushed the company toward lighter, more adaptable integration patterns. Rather than locking workflows into rigid custom code, architects now favor composable prompts and reusable prompt libraries that can be updated as models evolve. This approach reduces technical debt and keeps the focus on business process rather than infrastructure maintenance.

Data governance also surfaced as a critical workstream. Early experiments sometimes pulled information from sources whose freshness or permission levels had not been fully validated. The resulting hallucinations, though infrequent, damaged trust. Microsoft responded by building clear data lineage into every AI-assisted workflow. Users now see source citations alongside generated content, and sensitive fields are automatically masked or restricted. The governance layer did not slow adoption once teams understood it protected rather than hindered their work. On the contrary, confidence in the outputs increased, which encouraged even broader use.

The blog post also addresses the scaling challenge. What works for a nine-person product team does not automatically translate to a division of 3,000. Microsoft created a central transformation office that maintains a library of proven workflow patterns. Teams can browse patterns for sales forecasting, code review, contract analysis, or incident response and then adapt the closest match to their context. This approach prevents every group from starting from scratch while still allowing local ownership of the final design. The central team also tracks which patterns deliver the highest return on effort so that future investment can concentrate on the most promising areas.

Perhaps the most sobering lesson concerns the limits of AI on its own. Hogan repeatedly returns to the idea that technology alone does not change behavior. Without deliberate process engineering, the tools become faster ways to do the same old things. The 100,000 licenses distributed early in the program provided a useful on-ramp, yet they did not constitute transformation. Only when leaders accepted the harder work of redesign did measurable business value appear. That distinction, she argues, separates organizations that will capture AI’s full potential from those that will simply automate yesterday’s inefficiencies.

Microsoft continues to refine its approach. New experiments explore multi-agent systems that hand work between specialized models without human intervention at every step. Other teams are testing how AI can participate in strategic planning sessions by synthesizing market signals in real time. Each experiment follows the same discipline: begin with the business outcome, redesign the workflow that produces it, then embed AI where it adds the most leverage. The company has made the methodology available to customers through its advisory services, suggesting that the lessons learned internally now inform external guidance.

For organizations still early in their own AI efforts, the Microsoft experience offers a practical sequence. First, resist the temptation to declare victory based on license counts or monthly active users. Second, identify a handful of high-impact workflows where delay or error carries measurable cost. Third, assemble cross-functional teams to map those workflows in detail, highlighting every handoff, approval, and data translation step. Fourth, prototype new designs that place AI at the center rather than the periphery. Fifth, measure the revised process against the original baseline using the same key performance indicators that mattered before the tools arrived. Where the gap is large and sustained, scale. Where it is not, revisit the design.

The blog post ends on a note of cautious optimism. Microsoft has seen tangible improvements in speed, quality, and employee experience, yet leaders acknowledge that the work has only begun. New model capabilities will continue to arrive, and each wave will require fresh examination of existing processes. The organizations that embed continuous redesign into their operating rhythm, rather than treating it as a one-time project, stand the best chance of staying ahead.

By sharing both the successes and the stumbles, Microsoft has provided a reference point more valuable than any marketing campaign. The message is straightforward: licenses are easy, transformation is not. The difference between the two explains why some companies see only modest productivity gains while others, like the teams inside Microsoft that fully embraced workflow redesign, achieve step-change improvements in the metrics that matter most to their business.

from WebProNews https://ift.tt/qlMk7EW

AI Coding Tools Now Generate Over Half of Developers’ Code, Up 3X in a Year

The adoption of artificial intelligence tools among software developers has accelerated sharply over the past year, according to fresh data from BairesDev. The share of developers who now rely on AI to produce half or more of their code has climbed from 12 percent to 42 percent year over year. This finding comes from a survey of 705 developers and 41 enterprise chief technology officers conducted by the nearshore software development company. The results paint a picture of a profession in transition, where productivity gains exist alongside new responsibilities that keep total working hours from shrinking.

The survey, released in mid-September, offers one of the more detailed snapshots yet of how AI coding assistants are reshaping daily work. Respondents reported saving roughly 13 hours per week that would otherwise have gone to writing fresh code. Those reclaimed hours, however, have not translated into shorter workweeks or lighter workloads. Instead, developers are channeling the time into reviewing AI-generated output, debugging errors introduced by models, and mastering the rapidly changing array of tools at their disposal. Only 21 percent of developers still dedicate more than half their weekly hours to writing new code entirely from scratch. The rest have shifted toward verification, refinement, and integration tasks.

VentureBeat first highlighted the jump in AI-assisted coding volume, noting that the pattern holds across experience levels and company sizes. Developers with fewer than five years in the field showed slightly higher adoption rates, but even seasoned engineers reported incorporating AI into at least a quarter of their output. Overall, 72 percent of developers now use AI tools for 25 percent or more of their code, according to the same dataset recapped by Fudzilla.

The shift carries implications for both individual contributors and organizational strategy. As AI assumes a larger role in initial code generation, human attention concentrates on quality assurance. Sixty-seven percent of developers say they spend more time reviewing AI-produced code than they did a year ago. Fifty-two percent report increased debugging efforts aimed at issues that the models themselves introduced. These patterns suggest that while AI accelerates the creation phase, it simultaneously inflates the verification phase.

Chief technology officers appear to recognize this trade-off. Seventy-eight percent of the CTOs surveyed by BairesDev have increased spending on review, quality assurance, and validation processes. Rather than reducing headcount or budgets, many organizations are redirecting resources toward strengthening the human layer that sits atop automated generation. This includes hiring or training specialists in AI tool fluency, data engineering for model improvement, AI-specific quality assurance, and security practices tailored to machine-generated code.

The WebProNews analysis of the same BairesDev numbers framed the situation clearly: not one hour came back. Time saved on writing is consumed by oversight. Accountability, the article noted, remains firmly with the humans who ultimately ship the software. Developers cannot delegate liability for defects, security vulnerabilities, or performance problems to a language model. As a result, the profession is acquiring a new specialty—AI code reviewer—that demands skills in prompt engineering, output validation, and contextual understanding that models still lack.

This evolution raises questions about how software teams will measure productivity going forward. Traditional metrics focused on lines of code written or features delivered per sprint may become less relevant when half the lines arrive pre-written. Organizations may instead track defect density in AI-generated sections, time spent in code review, or the speed with which teams can safely integrate and deploy machine-assisted contributions. Some CTOs are already piloting new evaluation frameworks that weigh the accuracy of AI suggestions against the effort required to make them production-ready.

The survey also revealed regional and sectoral differences. North American teams reported slightly lower adoption rates than their Latin American and European counterparts, possibly reflecting stricter compliance requirements in regulated industries such as finance and healthcare. In contrast, startups and mid-sized technology firms showed the highest uptake, with many developers using multiple AI tools in parallel—GitHub Copilot for autocompletion, Claude for architectural suggestions, and specialized agents for test generation.

Training programs have begun to adapt. Several CTOs mentioned expanding internal academies to include modules on effective prompting, recognizing hallucinated code, and performing differential testing between AI and human implementations. Universities are starting to incorporate similar topics into computer science curricula, although the pace of change in industry continues to outstrip academic response.

Security concerns feature prominently in the findings. With AI models trained on public code repositories, the risk of introducing known vulnerabilities or inadvertently leaking proprietary patterns remains real. Thirty-eight percent of developers admitted they had discovered security flaws traceable to AI suggestions within the past six months. In response, many organizations now route all AI-generated code through static analysis tools and require human sign-off before merging into main branches. Some have implemented private model instances trained only on approved internal codebases, although this approach increases infrastructure costs.

The human factors dimension deserves equal attention. Developers who participated in the survey expressed mixed emotions about the technology. A majority appreciated the reduction in boilerplate work and the ability to explore alternative implementations quickly. Yet many also described a sense of cognitive overload when switching between writing, reviewing, and correcting AI output within the same hour. The context-switching tax appears to offset some of the raw productivity gains.

One senior engineer quoted in the Fudzilla coverage described the new workflow as “writing less but thinking more.” Instead of focusing on syntax, developers now concentrate on system behavior, edge cases, and alignment with business requirements. This shift may ultimately produce higher-quality software, provided teams can maintain focus amid the additional review burden.

Looking ahead, the BairesDev data suggests continued growth in AI coding assistance. Eighty-four percent of CTOs plan to increase investment in these tools over the next 12 months. The areas receiving funding include better integration with existing development environments, improved model specialization for domain-specific languages, and enhanced collaboration features that allow multiple engineers to critique the same AI suggestion in real time.

At the same time, expectations are becoming more realistic. Early hype around fully autonomous coding agents has given way to a more measured view that positions AI as a powerful junior pair programmer rather than a replacement for experienced staff. The survey found that only 9 percent of respondents believe AI will eliminate the need for human developers within five years. The overwhelming majority see the technology as augmenting rather than supplanting their roles.

This balanced perspective may prove healthy for the industry. By acknowledging both the time savings and the new obligations that accompany them, organizations can design workflows that play to the strengths of both humans and machines. Developers can focus on creative problem solving and architectural decisions while AI handles repetitive patterns and standard implementations. The key lies in building processes that keep humans in the loop without overwhelming them with review volume.

The BairesDev survey offers a useful benchmark for tracking progress. Future editions will likely explore how these patterns evolve as models improve in accuracy and as organizations refine their governance approaches. For now, the data confirms that AI has moved from experimental side project to core component of the modern development stack. Forty-two percent of developers generating at least half their code with AI represents a threshold that seemed distant only a year ago.

Companies that treat this transition strategically—investing in training, updating review practices, and measuring the right outcomes—stand to gain the most. Those that simply deploy tools without adjusting team structures or success metrics may find themselves spending the promised time savings on an expanded debugging backlog instead of innovation.

The story of AI in software development is still being written, but the latest chapter makes one outcome clear: the balance between generation and verification has shifted, and the humans who maintain ultimate responsibility are adapting their days accordingly. Whether this leads to shorter release cycles, fewer defects, or simply different kinds of work remains to be seen. What the numbers already show is that the profession is changing in measurable, concrete ways that demand attention from both practitioners and leaders.

from WebProNews https://ift.tt/8gPl2WE

Trump Launches ‘AI Force’ to Outpace China and Silence Safety Warnings

President Donald Trump took to Truth Social on Saturday with a blunt declaration. He is forming an “AI Force.” He will soon name an AI “czar.” Only “High I.Q. individuals need apply.”

The announcement lands as leading voices in the technology sector issue urgent calls to slow the breakneck pace of artificial intelligence development. Trump dismisses those concerns outright. He labels them the latest in a string of hoaxes pushed by political opponents. But the move reveals a calculated strategy. Speed up American innovation. Centralize oversight. Keep the U.S. ahead of China at all costs.

“Over the years, there have been many Hoaxes, all generated by the Radical Left Dumocrats, for purposes of destroying our Country,” Trump wrote. “And now, the decimation, or destruction, of AI, commonly known as Artificial Intelligence — And I, as President of the United States, will not stand by and let this happen.”

Short. Direct. Classic Trump. He compared the new entity to the Space Force he created during his first term. That initiative, he noted, proved a “tremendous SUCCESS.” The AI Force, he suggested, would protect the industry from interference while allowing it to flourish. “We will not in any way hinder or stifle the Growth of this incredible Industry. Rather, we will cherish it, help it, and watch over it, as it grows!”

Yet Trump added a caveat. The government would watch for misuse. “However, we will also be looking for BAD, and we can do that, very easily, with our already existing Criminal and Civil Justice System.” No new regulations. No heavy bureaucracy. Existing laws suffice. The message aligns with months of executive actions that prioritize rapid adoption across national security agencies.

From Executive Orders to a Dedicated Force

This latest declaration builds directly on earlier steps. In June, the White House issued a National Security Presidential Memorandum on Artificial Intelligence in the National Security Enterprise. It directed agencies to accelerate uptake of advanced models. Partnerships with commercial providers expanded. Talent pipelines strengthened. The goal was clear: give American warfighters and intelligence professionals decisive technological overmatch. (White House Fact Sheet)

An accompanying executive order from early June established a voluntary framework for frontier AI models. Developers could submit systems for government review before public release. The focus stayed on innovation paired with basic safeguards. Cybersecurity hardening. Protection of critical systems. Avoidance of any measures that might cede ground to Beijing. (White House Executive Order)

Trump’s Saturday post repeats themes from those documents. AI could represent as much as 25 percent of U.S. gross domestic product. It stands as the next Industrial Revolution. Larger than the internet. The United States leads China and the rest of the world. He intends to keep it that way. The race, in his view, leaves no room for hesitation.

Industry leaders see different risks. Executives at Anthropic and OpenAI recently warned about the dangers of unchecked progress toward superintelligence. They called for coordinated slowdowns. Trump responded sharply earlier in the week. He branded such fears a “SICK conspiracy.” He singled out Anthropic CEO Dario Amodei. The president claimed his administration already possesses “tremendous CRIMINAL and REGULATORY power over these companies.” (CNN Politics)

So the AI Force arrives at a tense moment. Details remain sparse. White House officials have not clarified whether it will function as a new military branch, a cross-agency task force, or something else entirely. The czar’s exact authorities stay undefined. Questions linger about staffing, reporting lines, and integration with existing bodies such as the Office of Science and Technology Policy.

But the intent shines through. Central coordination without regulatory drag. Private-sector talent injected into government operations. A structure that echoes the US Tech Force program announced late last year. That effort aimed to recruit roughly 1,000 AI engineers and specialists. It embedded them across departments including Defense and Treasury. Companies like AWS, Apple, and others signed on. (CBS News)

David Sacks, who previously served in a dual AI and crypto czar role, offered an earlier model. His tenure championed light-touch policies. The new czar will likely follow suit. Trump wants high-caliber minds. People who share his vision of unbridled American AI dominance.

Critics worry the approach leaves critical gaps. Public opinion polls show growing unease. A majority of Americans now view AI more as a potential harm than a benefit. Lawmakers from both parties have pushed for stronger guardrails on safety, bias, and job displacement. Trump’s stance puts him at odds with that sentiment heading into midterm elections.

And the international dimension complicates matters further. Treasury Secretary Scott Bessent met with Chinese counterparts this weekend. AI rivalry sits high on the agenda ahead of Trump’s upcoming summit with President Xi Jinping. Beijing advances its own models at a rapid clip. U.S. officials once described a multi-month lead. Industry voices now call it neck and neck. Export controls on advanced chips remain a flashpoint. Allegations of technology copying persist.

Trump’s announcement signals no retreat. Quite the opposite. The AI Force, in his framing, ensures the United States stays ahead. It counters what he sees as self-sabotage by domestic critics. Data center construction faced backlash over energy use and local impact. Trump likened that pushback to failed attacks on other priorities. The pattern, he argues, repeats with AI itself.

Supporters applaud the clarity. A single point of accountability. Fast decision-making. Reduced overlap between agencies. Skeptics counter that real risks demand more than existing statutes and a high-IQ appointee. Autonomous weapons. Misinformation at scale. Cybersecurity vulnerabilities in critical infrastructure. These issues have grown more pressing as models grow more capable.

The June national security memo addressed some of those points. It required annual reviews of autonomy guidance for weapon systems. It stressed clear chains of command. It barred any commercial entity from disabling AI tools that warfighters rely upon without approval. Practical measures. Yet they stop short of the comprehensive safety testing many experts advocate.

Trump’s latest move sidesteps that debate. He frames regulation itself as the threat. Slowing down hands advantage to adversaries. The AI Force exists to prevent that outcome. To watch over growth. To apply justice system tools where misuse appears. Simple in concept. Execution will prove far more complex.

Reactions poured in quickly on X. Some hailed the announcement as decisive leadership. Others questioned whether another layer of government coordination would accelerate anything. A few noted the irony. After years of complaints about bureaucratic bloat, Trump proposes yet another specialized body. But this one carries his personal brand. Modeled on a signature first-term achievement. Positioned as a bulwark against decline.

History offers mixed lessons. The original Space Force overcame early skepticism. It integrated space operations more effectively across military branches. Whether an AI Force can deliver similar focus remains untested. The technology evolves far faster than orbital mechanics. Models improve weekly. New capabilities emerge without warning. Any oversight structure must keep pace. Or risk irrelevance.

For now, the president has set the direction. Unleash AI. Appoint top talent. Maintain the lead. Details will follow. The choice of czar will signal priorities. Will the pick come from industry? Academia? National security circles? Trump has not said. He only set the bar. High intelligence. Alignment with his growth-first philosophy.

The coming weeks will test whether this approach satisfies lawmakers, reassures a wary public, or simply amplifies existing tensions. One thing appears certain. The U.S. government is doubling down on artificial intelligence as a core strategic asset. No slowdown. No apologies. Full speed ahead. The AI Force has been called into existence. Its real work begins when the czar takes the helm.

from WebProNews https://ift.tt/TgoUkD9

Fed Examiners Saw SVB’s Fatal Flaws a Year Early but Held Back, Fearing a Wrong Call

More than three years after Silicon Valley Bank collapsed in a spectacular March 2023 run, a new independent review has laid bare a striking truth. Federal Reserve supervisors knew or should have known about the bank’s deadly vulnerabilities as early as March 2022. They simply didn’t act with the force required.

The findings come from an outside examination commissioned by Michelle Bowman, the Fed’s vice chair for supervision. Starling Advisory Group conducted the work. Its conclusions, released Friday, paint a picture of paralysis born from fear. Examiners believed it safer to do nothing than risk getting a call wrong.

Bowman laid out the results in pointed remarks. “Our supervisory staff knew, or should have known, about these vulnerabilities as early as March 2022,” she said. Yet “supervisory staff did not take prompt and decisive action to encourage or require Silicon Valley Bank to reduce its interest rate risk or concentration of vulnerabilities.”

The vulnerabilities formed a toxic mix. Unrealized losses on the bank’s securities portfolio had wiped out its capital. Its deposit base stood at 94 percent uninsured and heavily concentrated among venture capital-backed technology companies. Management lacked operational readiness to borrow from the Fed’s discount window in a crisis. Any one issue might have been survivable. Together they proved fatal.

This account sharpens earlier post-mortems. The Fed’s own 2023 review led by then-Vice Chair Michael Barr criticized lax standards after 2018 regulatory easing and slow supervisory follow-through. It pointed to a less assertive culture under previous leadership. The new report goes further. It rejects the idea that the 2018 tailoring law or directives from prior officials caused the delays. The former vice chair for supervision had stepped down in October 2021, before SVB’s problems peaked.

Instead the Starling review homes in on something more insidious. A long-standing culture of risk aversion inside the supervisory ranks. Staff saw personal safety in inaction unless they held absolute certainty. Lack of clear decision rights only made things worse. No one knew exactly who could sign off on a tough call.

And social media? It played no meaningful role in accelerating the run, the report found. Analysis by Charles River Associates, brought in by Starling, showed 96 percent of relevant social media activity occurred only after the bank’s failure had become inevitable. The run started from real weaknesses, not online rumors.

Bowman didn’t mince words about the implications. “One significant factor contributing to supervisory inaction was a long-standing culture of risk aversion,” she said. “Staff believed it was personally safer to take no action unless they were certain the action was exactly right.” A lack of clarity on decision rights compounded the problem. Responsibility, authority and accountability had become disconnected across the system.

The review arrives at a delicate moment. Bowman, nominated by President Donald Trump, has already begun overhauling supervision practices. She plans staff reductions in the division. New supervisory operating principles stress earlier identification of threats and faster action. Examination teams must now file monthly reports to top leaders flagging any uncertainty about when or whether to act. The goal is real-time visibility and less fear of being second-guessed.

Critics wasted little time pushing back. Senator Elizabeth Warren called the report “an embarrassing attempt to re-write history designed to pave the way for more dangerous deregulation that will lead to the next Silicon Valley Bank disaster,” according to a Reuters article.

Her reaction reflects deep partisan divides over bank rules. The 2023 failure shook confidence in regional lenders. It forced emergency measures to backstop deposits and prevent contagion. No depositors ultimately lost money. But the episode exposed cracks in how midsize banks are watched.

Earlier analyses had reached similar ground. The Fed’s 2023 Barr report found supervisors failed to appreciate SVB’s risks as it ballooned from $71 billion to over $211 billion in assets between 2019 and 2021. It issued findings on governance, liquidity and interest-rate risk. Yet the pace remained deliberate. The bank held 31 open supervisory matters when it failed — three times the peer average. The Federal Reserve’s 2023 review called the approach too consensus-driven and slow.

A CEPR column from August 2026 went deeper. It argued the risks at SVB were visible for years. The bank had long funded long-duration securities with concentrated uninsured deposits. Supervisors focused on process compliance rather than forward-looking risk. They acted only once losses materialized amid rate hikes. The piece described supervision as policing process instead of actual exposures.

The Starling findings echo that view but assign clearer blame to internal caution. They also dismantle some prior excuses. Tailoring rules didn’t tie supervisors’ hands. No top-down order softened scrutiny. The problem sat inside the organization itself. Examiners hesitated because the personal cost of error felt higher than the institutional cost of delay.

Bowman insists the exercise isn’t about blame. “This review is not about assigning blame. Instead, it is about learning lessons from the past to avoid repeating them in the future,” she told her audience. The Fed has started addressing the culture head-on. Monthly escalation reports aim to surface doubts quickly. Leadership gains sightlines into gray areas. The hope is examiners will flag concerns without worrying about career repercussions.

Whether these steps will stick remains an open question. Banking supervision has always balanced judgment calls against second-guessing. Rate environments shift. Business models evolve. Concentrated deposit bases can vanish overnight, as SVB proved when its tech clients pulled funds en masse.

The report lands the same week the Fed raised interest rates for the first time since 2023. Higher rates amplified SVB’s unrealized losses in 2022. Today’s environment carries different pressures. Commercial real estate exposures, for instance, have drawn fresh scrutiny at other regional players.

Industry insiders have watched the regulatory pendulum swing for decades. Post-2008 rules tightened dramatically. The 2018 Economic Growth, Regulatory Relief, and Consumer Protection Act dialed some back for smaller institutions. SVB sat right at the edge of heightened standards. Its growth outran the gradual phase-in.

Yet the new review suggests the real failure wasn’t in the rulebook. It was in execution. Supervisors saw the problems. They documented them. They just couldn’t pull the trigger fast enough. Certainty became the enemy of timeliness.

Bowman has signaled broader changes ahead. Reduced headcount in supervision. Clearer principles. Faster escalation paths. The test will come in the next stress point. Will examiners act on early warnings, or will the same risk aversion reassert itself?

SVB’s collapse didn’t topple the system. Swift government intervention contained the damage. But the second-largest bank failure in U.S. history left a mark. It showed how quickly confidence can evaporate when uninsured depositors smell trouble. It revealed gaps in liquidity planning and interest-rate hedging at institutions that seemed sophisticated.

The Starling report won’t end the debate. Warren and others see it as cover for loosening rules. Supporters of Bowman’s approach view it as a clear-eyed diagnosis free from prior political lenses. What matters most is whether the cultural fixes take hold.

Because the next time a bank sits on large unrealized losses and a flighty deposit base, supervisors will face the same choice. Act early and risk being called heavy-handed. Wait and risk another slow-motion disaster. The new procedures aim to make that choice less fraught. History suggests it won’t be easy.

Friday’s release adds one more layer to the SVB story. It doesn’t rewrite the facts of the bank’s mismanagement. SVB’s own leadership failed to grasp or address its exposures. But it does sharpen accountability for the watchdogs. They saw it coming. They knew enough. They held back anyway. The reason, according to this latest account, was simple. They feared being wrong.

from WebProNews https://ift.tt/OSraHYy

How AI Powers an Unchecked Theft Machine

Matt Stoller didn’t mince words. In his newsletter published today, the antitrust advocate laid out a blunt case: the rush to build ever-larger AI systems rests on systematic violations of law that regulators have chosen not to pursue. The Big Newsletter called it an elite crime spree. Documents unsealed in the New York Times copyright suit against OpenAI show the company circumvented paywalls to scrape content. When informed of the hack, President Greg Brockman replied, “ah nice.”

A Microsoft director of applied science went further. He described training big models on copyrighted material as “the largest theft of labor in human history.” Those words come from court filings uncovered by Jason Kint. They suggest possible breaches of the Computer Fraud and Abuse Act. Yet enforcement remains absent.

Short. Direct. The pattern repeats across the industry.

While tech executives warn of future existential risks, present-day harms multiply. Voice clones mimic grandchildren in distress. Deepfakes pose as executives demanding urgent wire transfers. Synthetic identities slip past verification systems at scale. Losses have reached hundreds of millions in the United States alone. And the technology that enables this arrives cheaper and more convincing each quarter.

The FBI’s Internet Crime Complaint Center tracked the shift. In 2025 it received more than 22,000 complaints tied to AI tools. Reported losses hit $893 million. Investment scams accounted for $632 million of that total. Older Americans bore a heavy share, losing $352 million. Yahoo News reported the figures yesterday.

But those numbers capture only what victims recognize and report. Many scams now arrive polished. A few seconds of audio from social media suffices for convincing voice clones. AI writing tools generate error-free phishing messages that once betrayed their origins through awkward phrasing. The Washington Post explained the change this week. Scams aren’t new. Their cost and quality have transformed.

Organized networks recycle the same fraudulent assets across targets.

Shufti’s 2026 Identity Fraud Report, released last week, paints an even darker picture. Deepfake document fraud made up 80 percent of AI-enabled attacks in the first half of the year. Synthetic identities followed at 12 percent. The firm projects a 495 percent rise in AI-powered identity fraud for 2026 compared with 2025. Organized rings reuse the same forged documents and devices. One network linked 70 identities through just 13 devices. Digital Watch Observatory covered the report on September 11.

TRM Labs tracked parallel growth in crypto-related crime. Its 2026 AI-in-Crime Adoption Index climbed to 54 from 28 in 2024. Deepfake-scam losses in early 2026 already exceeded the entire previous year by 263 percent. Deloitte forecasts U.S. generative-AI-driven fraud losses will reach $40 billion by 2027, up from $12.3 billion in 2023. The firm published its analysis this month. TRM Labs detailed the index on September 17.

And. The same tools power romance scams that once required weeks of patient grooming. Incode’s Agentic Fraud Report documented 66 incidents, 44 of them confirmed AI cases. Autonomous agents now handle target research, conversation maintenance and fund extraction with minimal human oversight. Global fraud losses reached $579 billion in 2025 according to Nasdaq Verafin data cited in the report. Incode released its findings today.

Real people feel the impact. A Georgia man lost his car and cash after months of communication with an AI-generated persona posing as a sheriff’s deputy. Police arrested 25-year-old Caleb Mills on charges including identity fraud and impersonation. He allegedly used stolen photos, fake accounts and voice-changing tools. Cybernews reported the case on September 9.

Another victim, Kris Kolakosis, handed over $400,000 in a catfishing scheme built on AI-generated profiles and conversations. He met “Eliza” on Facebook. The interaction felt genuine until the requests for money began. Newsgram told his story in July.

Corporations face sophisticated variants. In one documented case, fraudsters used a cloned executive voice during a video call to authorize a large transfer. Such business email compromise schemes involving AI generated $30 million in reported FBI losses for 2025. The bureau has warned repeatedly about deepfakes impersonating government officials, including videos of senior FBI agents directing victims to fake recovery sites.

But what about the foundation? The training data itself.

Stoller argues the entire edifice depends on mass copyright infringement and unauthorized access. Hyperscalers scrape the internet without permission. They circumvent technical protections. When caught, responses range from indifference to celebration. Existing laws already prohibit these acts. The Computer Fraud and Abuse Act. Copyright statutes. Antitrust rules against illegal monopolies. Yet prosecutors rarely charge the powerful.

Sam Bankman-Fried once served as a major early backer of Anthropic. Meta faced accusations of facilitating mass sex trafficking on its platforms. Financial maneuvers fund the enormous data centers required. Each element adds to a picture of an industry operating beyond normal legal constraints.

Enforcement lags for a reason. Political pressure favors new regulations over application of old ones. Safety standards modeled on the FDA gain traction among some lawmakers and even certain AI firms. Bernie Sanders calls for oversight. Anthropic, OpenAI and Google support versions of it. Stoller counters that such measures miss the point. The problem isn’t lack of rules. It’s selective blindness to violations by elites.

Recent actions show tentative pushback. The Manhattan District Attorney seized 12 domains selling AI-generated non-consensual deepfake pornography in mid-September. The sites allegedly turned photos of 1,200 real people into explicit videos without consent. Victims included actors, politicians and influencers. Manhattan DA’s Office announced the seizures on September 14.

Still, these represent surface-level responses. The deeper architecture remains untouched. Organized crime networks in Southeast Asia use American AI models like ChatGPT and Gemini to automate multilingual scam operations. One tool set generated tens of millions in illicit profits. AP and FRONTLINE documented the supply chain in June. Scammers pay for specialized software built on U.S. tech. The profits flow back through crypto wallets.

Projections point higher. Incode estimates AI could accelerate global scam losses toward $1 trillion annually if trends continue. Shufti sees fraud rings scaling through reusable synthetic assets. TRM Labs notes the industrialization of deepfake production across dozens of countries.

So the panic over future AI dangers feels oddly timed. Current capabilities already extract value at unprecedented scale. They steal creative labor. They defraud individuals and businesses. They erode trust in voice, image and video. All while the architects claim to race toward beneficial artificial general intelligence.

Critics like Stoller demand something simpler. Enforce the laws on the books. Treat the “largest theft of labor in human history” as a crime worth prosecuting. Apply antitrust standards to break illegal monopolies. Pursue clear violations of computer fraud statutes.

Without that shift, new regulations risk becoming theater. They may slow responsible developers while sophisticated operators continue unchecked. The crime spree, elite in its participants and global in its victims, rolls on. Billions in losses accumulate. And the public grows weary of promises that tomorrow’s safeguards will fix today’s thefts.

The evidence sits in court documents. FBI reports. Industry analyses released this month. The pattern is clear. AI doesn’t just enable fraud. In many corners it was built on it.

from WebProNews https://ift.tt/MIpltZG

Foreign Central Banks Turn Away From U.S. Treasuries as Holdings Hit Multi-Year Lows

Foreign central banks and governments have soured on U.S. Treasury securities. Their holdings slid again in July. Official foreign owners now sit on just $3.77 trillion in Treasuries at market value. That’s roughly the same dollar amount they held back in 2012.

But context matters. The total stock of marketable Treasuries has tripled since then. Inflation has added 48% to the price level over that span. The share of Treasuries in foreign official hands has plunged to 12.8%. Not seen since 1993. The shift leaves the U.S. far less reliant on these traditional buyers.

Latest data from the Treasury Department’s TIC report show total foreign holdings of U.S. government debt dropped $50.4 billion in July to $9.25 trillion. The lowest mark since October 2025. (Wolf Street, Sept. 17, 2026).

Japan cut its position by $13 billion. That brings its reduction since February to $135 billion. Tokyo needed dollars to defend the yen. Multiple intervention rounds explain the sales. Don’t expect tears from Japanese authorities. Those operations have proven profitable.

China’s official holdings fell another $15.4 billion to $618 billion. An 18-year low. The figure matches levels last seen in August 2008. Combine mainland China and Hong Kong. The pair shed $13 billion in July alone. Over the past 12 months the decline totals $67 billion. The long retreat from the 2013 peak above $1.3 trillion continues without pause.

France saw a sharp drop of $41.5 billion. Its holdings stand at $348.4 billion. Canada trimmed $33.3 billion. Leaving it with $426.3 billion. The UK moved the other way. Adding $58.4 billion to reach $998.3 billion. Still, the net picture shows official sellers dominating.

These moves reflect deeper forces. Central banks hunt alternatives. Gold stands out. China’s reserves have shifted noticeably toward the metal. Over the past decade Beijing cut Treasury exposure by 47% while lifting gold holdings 27%. India took the opposite path. Raising both Treasuries and gold.

Wei Li, head of multi-asset investments at BNP Paribas Securities in China, points to a clear pattern. “A global trend of diversification into gold and agency bonds, as well as other assets like equities, especially with the AI boom.” (Financial Times, Sept. 17, 2026). Real yields sit near highs not seen since 2008. Yet the old negative link to gold prices broke in 2022. Central banks keep buying the metal anyway. The World Gold Council reports 22 straight months of net accumulation. Reserves now top $5 trillion. Surpassing foreign official Treasury holdings.

Geopolitics adds fuel. Beijing worries about asset freezes. Russia’s experience in 2022 lingers. U.S. fiscal deficits swell. Inflation risks persist. Trade surpluses no longer flow so readily into American paper. The two economies face opposite pressures. America wrestles with deficits and price pressures. China battles slowing growth and deflation.

But the story isn’t simple flight. Private foreign investors keep buying. They pushed total foreign holdings higher in recent years even as officials sold. Opaque financial centers lead the charge. Belgium holds $471 billion. Cayman Islands $460 billion. Luxembourg $442 billion. Ireland $350 billion. Switzerland $285 billion. Singapore $278 billion. Many of these positions reflect U.S. hedge funds and corporations parking money offshore. The basis trade lives here. So does corporate America with overseas entities.

Foreign official holdings now make up only 41% of overseas Treasury ownership. Down from two-thirds in 2014. The absolute level sits just 8% below 2014 peaks. Yet the overall Treasury market tripled. Reserve accumulation slowed globally. The Fed’s own balance sheet expansion absorbed supply. Dollar strength forced rebalancing.

Recent reports confirm the trend. China’s July figure marks the lowest since 2008. France and Canada led the monthly drop. UK buying provided only partial offset. (The Nation Thailand, Sept. 18, 2026). Analysts note China’s true exposure may exceed reported numbers. Custodial holdings in Belgium and Luxembourg obscure part of the picture.

Implications stretch wide. The U.S. must court different buyers to fund its deficits. Yields rise to attract them. The 20-year auction last week cleared at 5.42%. Indirect bidders showed limited appetite. Private leveraged players fill gaps. But their commitment differs from patient central bank money.

Gold’s surge tells part of the tale. Central banks accumulated 1,000 tonnes annually in recent years. Double the prior decade’s pace. 89% of respondents in a June World Gold Council survey expect further gains ahead. The old correlation with real yields no longer holds. Gold trades near $4,300 an ounce. Far above levels implied by historical relationships.

Japan’s sales carry special weight. Its interventions drained foreign currency reserves by $95 billion in August. Securities accounted for most of the drop. Yet Tokyo turned a profit overall. Future defenses may rely more on borrowing arrangements at the Fed rather than outright sales.

The numbers don’t lie. Official foreign ownership share collapsed from 34% in 2012 and over 38% at the 2007-2009 peak. To 12.8% now. Total foreign ownership hovers near 31-34% of the market. Stable on the surface. But the composition changed. Private capital. Often leveraged. Often domiciled in Caribbean or European financial centers. Often tied to U.S. entities.

This evolution carries risks. Basis trades can unwind fast if funding costs spike. Private money proves fickle in stress. Central banks once provided stable demand. Their absence forces higher yields or bigger Fed involvement. Or both.

Recent TIC data paint a consistent picture. Declines in July followed similar moves in prior months. Japan. China. France. Canada. All trimming. The UK stands as notable exception. India added modestly in some periods but shows longer-term caution. Brazil and others vary.

Broader research supports the view. Slower global reserve buildup explains much of the shift. Fed holdings. Currency rebalancing. Geoeconomic fragmentation reduces demand from countries distant from U.S. policy. Private demand remains more sensitive to safe-haven flows. (Bloomberg, Sept. 16, 2026).

The U.S. debt stands at $40 trillion and climbing. New supply floods the market. One trillion dollars absorbed in recent three-month periods. Buyers demand compensation. Yields adjust. The old comfortable reliance on foreign official capital has faded. A new balance emerges. One that depends more on domestic buyers, private foreign capital, and market-driven rates.

And that change won’t reverse soon. Diversification continues. Gold buying persists. Fiscal pressures in Washington show no sign of easing. Foreign central banks have found Treasuries less appetizing. The market adapts. But not without higher costs.

from WebProNews https://ift.tt/KLcu7kg

Bitcoin’s Uncertain Path: Why $100,000 Hopes Clash With Fresh Setbacks

Bitcoin trades near $76,000 this week. The cryptocurrency has shed gains from August’s rally. Fresh pressure comes from failed legislation and signals of tighter monetary policy ahead. Investors who bet on steady institutional inflows now face renewed questions about near-term direction.

The setback arrived Tuesday when the U.S. Senate blocked advancement of the Digital Asset Market Clarity Act. The procedural vote fell short at 49-50. That bill had promised clearer rules for digital assets. Its stall removes a potential catalyst. Bitcoin dropped below $75,000 shortly after the news broke. Moneycontrol reported the move alongside rising Treasury yields and oil prices that rattled risk assets broadly.

Yet this isn’t the first time Bitcoin has absorbed bad headlines. Spot Bitcoin ETFs recorded more than $450 million in outflows on Sept. 15. That marked the largest single-day redemption since June. BlackRock’s iShares Bitcoin Trust and Fidelity’s fund led the exits. Still, the products have pulled in tens of billions since their 2024 launch. Demand hasn’t vanished. It has simply turned fickle.

Analysts remain split. Some see a base case around $80,000 by year-end. Others warn of a retest near $70,000 or lower. The Yahoo Finance piece from earlier this year laid out scenarios that still resonate. It highlighted how macroeconomic factors and institutional adoption could drive prices higher. But it also flagged risks from regulation and market cycles. Yahoo Finance explored whether Bitcoin could reach fresh highs before the fourth quarter. Current trading suggests patience is required.

Bank forecasts vary widely. Standard Chartered’s Geoffrey Kendrick targets $100,000 by December. He has dialed back earlier optimism but still sees structural buying from ETFs. Bernstein projects $150,000 by late 2026. The firm argues Bitcoin has moved past its traditional four-year cycle. JPMorgan goes further. It sees $170,000 possible if Bitcoin starts to mirror gold’s role in portfolios. These calls assume continued capital allocation by institutions.

Bearish voices push back. NYDIG, Citigroup and Fidelity sketch ranges from $38,000 to $75,000. They cite sticky inflation, potential Federal Reserve tightening and the lingering pull of the four-year cycle. CryptoSlate’s September model puts the median terminal price near $88,000 by mid-December. Its bullish case reaches $114,000 while the bearish scenario lands at $67,000. Black-swan stress tests point even lower. CryptoSlate updates these figures regularly based on market data.

September has brought mixed signals. Bitcoin rose nearly 25 percent in August. It briefly topped $82,000. Support held near $76,000 for much of the month. But the recent break below that level activates technical concerns. On-chain data from Glassnode shows heavy supply between $81,000 and $86,000. Whales accumulated 39,000 BTC worth roughly $3 billion in late August. That buying helped stabilize price then. Whether it returns now is unclear.

Short-term holders felt the pain this week. Exchange inflows from newer investors spiked. More than 23,000 BTC moved at a loss. It was the largest such capitulation event in September. Yet many shook it off quickly. Long-term holders continue to hodl. Their behavior has supported Bitcoin through previous drawdowns.

Federal Reserve policy looms large. Markets assign over 90 percent odds to a 25-basis-point rate hike at the September meeting. Chairman Kevin Warsh faces higher Treasury yields. The 10-year note crossed 5 percent this week for the first time since 2023. Global bond yields have hit multidecade highs in several economies. Higher rates typically weigh on speculative assets. Bitcoin is no exception.

And then there is the ETF story. Inflows rebounded modestly mid-month before the latest outflows. BlackRock and Fidelity captured most of the positive flows when they occurred. Grayscale’s GBTC continues to see redemptions. Concentration in a few funds highlights how institutional participation remains uneven. Cryptonomist noted the lopsided nature of recent activity on Sept. 14.

Prediction markets add another layer. Polymarket users give low odds for the Clarity Act becoming law this year. They price meaningful chances for Bitcoin both above $80,000 and back toward $70,000 in September. Traders appear prepared for volatility. Some models from AI platforms like Claude and Gemini point to ranges between $75,000 and $97,000. ChatGPT assigns the highest probability to a $74,000-$88,000 band for the month.

Technical levels matter now. A weekly close above $77,100 would ease immediate pressure. Failure to hold $73,500 opens the door to $70,000. That zone aligns with cost basis for holders of three to six months. Support at the bull market band sits near $70,000 as well. Breaking lower would test the cycle-low thesis that some analysts still defend.

Bitcoin’s history shows resilience. It has recovered from deeper drawdowns. The 2022 bear market took it below $20,000 before the current expansion. Institutional infrastructure built since then changes the equation. ETFs provide easier access. Public companies add Bitcoin to balance sheets. Miners pivot toward artificial intelligence deals worth billions even if revenue remains thin. CryptoSlate highlighted those AI-related mining developments this week.

Yet risks abound. Regulatory clarity remains elusive after the Senate vote. Macro conditions could tighten further if inflation data surprises to the upside. Oil above $80 per barrel adds to those pressures. Short-term traders face liquidations on both sides of the market. Leverage remains elevated in derivatives.

So what comes next? Bank targets cluster between $100,000 and $170,000 for 2026 under bullish assumptions. More conservative models see an average near $72,000 for the year. The gap reflects genuine uncertainty. Bitcoin no longer moves in isolation. It reacts to bond yields, Fed decisions and legislative outcomes in real time.

Investors who entered during the 2024-2025 run-up sit on varied returns. Those who bought the October 2025 peak near $126,000 remain underwater. Others who accumulated below $70,000 this summer hold gains. The dispersion in outcomes mirrors the dispersion in forecasts.

Market participants watch the Fed announcement closely. A hawkish tone could extend the current pullback. A surprise pause might spark relief buying. Either way, the path to higher prices looks bumpier than many expected months ago. Bitcoin has traded in a broad range for most of 2026. Breaking out will require sustained demand that overcomes these headwinds.

History suggests it eventually does. The question is when. And at what price. For now the market digests fresh losses and waits for the next clear signal.

from WebProNews https://ift.tt/k7W3hgc

Atlassian Launches Agent Loops: AI Agents Automate Full Software Development in Jira

Atlassian has introduced a coordinated system for agentic engineering that connects autonomous AI agents across the software development lifecycle inside Jira. The new capability, called agent loops, allows a backlog item to flow automatically from planning through code generation, testing, and into a reviewed pull request without constant human prompting at every step.

The system relies on what Atlassian calls the Teamwork Graph, a live knowledge structure that maps how teams actually work. It pulls together information from Jira issues, Confluence pages, Bitbucket repositories, Compass components, and historical pull request patterns. When an agent begins work on a ticket, it first consults this graph to understand the surrounding context: which services are affected, how similar changes were implemented before, who owns related code, and what acceptance criteria mattered in past comparable stories. That shared memory reduces the common problem of agents producing code that looks correct in isolation but conflicts with team standards or architecture decisions.

Once the loop starts, a product agent interprets the backlog item and breaks it into granular tasks. A coding agent then generates implementation, drawing directly from the Teamwork Graph to match existing patterns, naming conventions, and error-handling approaches already used in the codebase. The generated code moves to an automated testing agent that writes and runs unit, integration, and contract tests. If failures appear, the loop returns control to the coding agent with specific failure logs and reattempts the fix. Only when all tests pass does the system open a pull request, complete with a detailed explanation of changes, links to related tickets, and suggested reviewers based on ownership data in the graph.

Atlassian demonstrated the loop handling an end-to-end feature addition in a sample microservices application. The entire cycle from backlog item to merged pull request completed with minimal engineer intervention beyond final approval. Early internal testing showed that loops can reduce cycle time on well-scoped tickets by more than half compared with traditional developer workflows, although results vary by complexity and team maturity.

Alongside the agent loops announcement, Atlassian released new developer experience measurement tools integrated into Jira and Compass. These tools track a set of signals that research has connected to actual engineering output: deployment frequency, lead time for changes, change failure rate, and time spent in context switching. The measurements go beyond simple velocity metrics by incorporating qualitative signals such as satisfaction scores from periodic developer surveys and the volume of blocked time logged against tickets.

The DX analysis accompanying the launch examined data from hundreds of organizations using Atlassian platforms. Teams whose members maintained the highest density of connections inside the Teamwork Graph—meaning they actively updated documentation, linked related issues, and kept component ownership information current—shipped approximately 64 percent more features per developer over a six-month period than teams with lower graph density. The correlation held after controlling for team size, domain complexity, and engineering tenure. Atlassian presents the finding as evidence that shared organizational memory becomes even more valuable when AI agents can read and act on that memory at scale.

The company positioned the combined release as a practical step toward what it calls coordinated agentic engineering rather than isolated point tools. Instead of developers copying context into separate AI chat windows, the system keeps every agent operating inside the same information environment that human teams already use. Jira remains the single source of truth: agents create sub-tasks, update status, attach artifacts, and log decisions without requiring users to switch applications.

Security and governance received attention in the design. All agents operate under the same permission model as the human user who triggered the loop. An agent cannot access repositories or services that the triggering engineer could not reach. Administrators can set organization-wide policies that require human review before any generated code reaches production branches. Audit logs capture every decision an agent makes, including which parts of the Teamwork Graph it consulted and why it chose particular implementation approaches.

Early adopters include both established enterprises and smaller product teams. One financial services company reported using loops to handle routine API extensions, freeing senior engineers to focus on complex regulatory requirements that still demand deep human judgment. A gaming studio adopted the system for repetitive gameplay feature scaffolding, noting that the generated code consistently followed the studio’s established architectural patterns because the Teamwork Graph captured those patterns accurately.

Atlassian built the loops on top of its existing Forge platform, which allows the agents to run in a secure, isolated environment close to customer data. The architecture avoids sending sensitive code or proprietary business logic to external large language model providers unless customers explicitly choose to route certain workloads through models hosted by OpenAI, Anthropic, or Google. Most processing happens inside models fine-tuned by Atlassian on anonymized public code and internal best-practice patterns.

The DX measurement tools introduce a new dashboard inside Jira that visualizes four key quadrants: throughput, stability, satisfaction, and cognitive load. Throughput combines deployment frequency with story points completed per engineer. Stability tracks failure rates and mean time to recovery. Satisfaction pulls from short pulse surveys sent through Jira. Cognitive load measures time spent waiting for builds, attending unplanned meetings, and chasing context across tools. The system can segment these metrics by team, product area, or individual contributor, allowing engineering leaders to identify specific bottlenecks.

One notable pattern in the DX data involves the relationship between documentation hygiene and delivery speed. Teams that kept their component catalog in Compass up to date and maintained clear links between Jira tickets and architectural decision records showed both higher agent success rates and better human developer throughput. When the Teamwork Graph contains stale information, agents generate more incorrect assumptions, leading to increased pull request rework. The 64 percent productivity difference appeared most strongly in organizations that treated knowledge maintenance as an ongoing engineering responsibility rather than a documentation afterthought.

Atlassian plans to expand agent capabilities in coming quarters. Future loops will include design agents that can propose user interface changes consistent with existing design systems, documentation agents that update Confluence pages in parallel with code changes, and release agents that coordinate rollout plans across multiple services. Each new agent type will continue to consult the same Teamwork Graph, creating a growing network of specialized workers that share context without duplication.

The company also announced an open beta for custom agent development. Teams can define their own loop patterns using a visual workflow builder inside Jira. A retail company, for example, might create a specialized loop that automatically generates new product catalog endpoints whenever a business analyst adds an item to the merchandising backlog. Because the custom loop references the Teamwork Graph, the generated code automatically includes the correct pricing service integration patterns already used by that organization.

Feedback from the initial pilot programs highlighted both strengths and limitations. Engineers appreciated that the generated pull requests included clear explanations and linked directly to original requirements, making review faster. However, participants noted that highly novel features or those involving significant refactoring still required substantial human guidance. The loops perform best on work that has clear precedents inside the Teamwork Graph. As organizations expand their graph with more historical decisions and architectural patterns, the range of tasks that agents can handle autonomously is expected to grow.

Pricing for the agent loops follows Atlassian’s existing model of per-user subscriptions with additional consumption charges based on the number of loop executions. Organizations already on Premium or Enterprise plans for Jira and Bitbucket receive basic loop functionality at no extra cost during the initial rollout period. Advanced DX analytics and custom agent development carry separate add-on fees.

The launch reflects Atlassian’s broader strategy of embedding intelligence directly into the tools where engineering work already happens. Rather than asking developers to adopt new AI platforms, the company brings agent capabilities into the familiar Jira interface and ties them to the organizational memory already captured in its products. The combination of autonomous loops and quantitative developer experience measurement gives both individual contributors and engineering leaders concrete data about where AI assistance accelerates delivery and where human expertise remains essential.

As more organizations begin experimenting with these coordinated agents, the industry will gather evidence about which types of software work benefit most from autonomous execution and which still require close human oversight. Atlassian’s early data suggests that the quality of shared team knowledge may prove as important as the sophistication of the underlying models. Teams that invest in maintaining accurate, connected information appear positioned to gain the largest advantage as agent loops become a standard part of the development process.

The release marks a concrete milestone in moving AI assistance from individual productivity aids toward coordinated systems that can carry out multi-step engineering workflows. By grounding those workflows in the actual collaboration patterns of each organization, captured in the Teamwork Graph, Atlassian aims to make agentic development feel like an extension of existing team practices rather than a replacement for them. The accompanying DX measurement tools provide organizations with visibility into whether those agents are delivering measurable improvements in both speed and developer satisfaction.

Over the next year, continued refinement of the loops, expansion of available agent types, and deeper integration with the measurement dashboard should give engineering organizations richer options for deciding which work to assign to autonomous systems and which to keep in human hands. The 64 percent productivity correlation observed in the initial analysis offers an early indicator that attention to organizational memory may become a competitive differentiator in an environment where AI agents can read and act on that memory at scale.

from WebProNews https://ift.tt/nec8Gwa

Cops Type ‘LMAO’ and Keyboard Mash to Query Vast License Plate Camera Networks

A deputy in Lake County, Indiana, once scanned a license plate across more than 19,000 cameras spanning 1,558 cities and towns. His official reason for the query? “LMAO.”

That single entry captures a pattern now documented across dozens of police agencies. Officers have filled the mandatory “reason” field in Flock Safety’s surveillance system with jokes, insults, shrugs and random keystrokes. The revelations come from an analysis of audit logs obtained through public records requests.

The Electronic Frontier Foundation reviewed the data and found entries such as “LOL,” “Hehe,” “Haha,” “idk,” “blah,” “TBD” and the longer plea “robbery I don’t remember the case number leave me alone.” Some officers typed insults like “idiot,” “dickhead,” “shithead,” “fuck this new search engine” or “WEIRD KID.” Others simply mashed the keyboard: “asdfg,” “gyghkkghghjkghjk,” “jhjhjkhj,” “nmbvcbnm.”

These searches took place from 2023 through late 2025. Flock altered its system afterward, making such logs harder to obtain through open records. Yet the entries expose something deeper than sloppy typing. They reveal a casual attitude toward a tool that tracks vehicle movements on a massive scale.

Flock Safety has built one of the largest automated license plate reader networks in the United States. Its cameras capture images of passing cars, extract plate numbers and store location data. Police can query the system for specific plates or broader patterns. The company markets the technology as a crime-fighting asset. But the audit logs tell a different story. And they arrive at a moment when Flock faces growing scrutiny.

Last month, USA TODAY published an investigation that examined more than 75 million license plate searches conducted by officers from 6,300 agencies between January 2023 and July 2026. The analysis uncovered patterns of potential misuse. At least 19 police and sheriff’s office employees in five states were arrested, fired, placed on administrative leave or investigated in the following weeks for improper use of the system.

One case involved a Georgia officer who queried her estranged husband’s plate almost daily for months, sometimes dozens of times in a single day. Another Texas detective searched more than 80,000 images without a warrant while investigating a suspected abortion. These examples sit alongside the frivolous entries uncovered by the EFF.

But the problem runs larger than isolated jokes. In a separate dataset of 11.4 million nationwide Flock searches over six months, the ACLU determined that more than 14 percent listed only the word “investigation,” with no case number or further detail. Similar vagueness appears repeatedly. Officers write “test,” a single letter or nothing at all.

The scale matters. A single query can pull data from thousands of cameras across hundreds of jurisdictions. One Goshen, Indiana, officer searched 6,474 automated license plate reader networks — representing 82,413 cameras — and entered “idk” as the reason. Harris County, Texas, sheriff’s employees used “LOL” or “lol” as a case number on multiple occasions. Kankakee County, Illinois, officers chose “idk” or “idk lol.”

Eatonton, Georgia, police entered strings such as “HJKNUILH,” “uiokjk.kuj” and “GJLHBNMN.” An Atlanta officer used “asdfga.” These are not outliers. The EFF identified button-mashing as its own category of search behavior.

Such practices raise basic questions about accountability. The “reason” field exists to create a record of legitimate law enforcement need. When that field contains gibberish, the record becomes meaningless. Privacy advocates argue the entire system operates with insufficient oversight. Warrants are rarely required. Retention policies vary. Data sharing crosses jurisdictional lines with little transparency.

Flock has responded to criticism by pointing to its value in solving serious crimes. The company notes that its cameras have helped recover stolen vehicles, locate missing persons and support investigations into violent offenses. Executives have said the technology gives police a tool they would otherwise lack. Yet the same system that catches real offenders also enables casual browsing, personal vendettas and outright abuse.

Recent news shows the tension playing out in real time. On the same day the EFF findings appeared, Atlanta officials announced plans to introduce citywide surveillance rules. An internal police audit of 115,578 Flock inquiries flagged 79 searches for further review. Mayor Andre Dickens stated that “technology that gives government extraordinary capabilities must come with extraordinary accountability.” The proposed ordinance would make internal police safeguards legally binding across all city departments.

Public backlash has taken other forms. In College Station, Texas, a 19-year-old man cut power cables to two Flock cameras and covered them with American flags. He now faces a criminal mischief charge. Similar vandalism has occurred elsewhere as residents grow uneasy with the expanding camera presence. Flock itself reports that its network recently surpassed 100,000 cameras nationwide, even as more than 50 cities have canceled contracts over concerns about data sharing with federal agencies.

The company adjusted its systems in December 2025 after increased public attention. Those changes made certain audit logs less accessible through records requests. The timing aligns with the end of the period examined by the EFF. Critics see the move as an effort to reduce transparency rather than address underlying problems.

Earlier reporting added context. The Houston Chronicle documented how Houston police conducted thousands of searches with vague justifications such as “investigation,” “suspect” or simple gibberish like “asdf.” Usage grew rapidly. Weekly searches doubled in less than a year. Some officers admitted to running “random” checks while on patrol.

WIRED examined Flock’s newer AI-powered tools. The systems now allow officers to search not just by plate but by descriptive prompts. An officer can ask for vehicles or people matching certain characteristics within a drawn geographic area. The technology stops short of facial recognition but edges closer to broader behavioral tracking. Privacy researchers worry that weak controls on the “reason” field become even more consequential when queries grow more sophisticated.

The EFF’s report lands amid a wider debate. Some lawmakers push for stricter rules. Colorado recently required warrants before sharing automated license plate data with federal agencies. Other states consider similar measures. Police departments defend the technology as essential for public safety. They argue that focusing on misuse ignores the many legitimate successes.

Yet the logs keep surfacing. Insults directed at suspects or the system itself. Expressions of ignorance or indifference. Pure nonsense. Each entry represents a moment when an officer decided that documenting a real justification was unnecessary. Or perhaps too much trouble.

That indifference carries consequences. Every plate captured contributes to a detailed map of movement. Over time, those maps can reveal where people live, work, worship or seek medical care. When the system treats such data casually, trust erodes. Residents already express discomfort with cameras that never sleep. The joke entries only heighten that unease.

Flock continues to expand. New contracts outpace cancellations. The network grows larger and more interconnected. At the same time, journalists, activists and some officials dig deeper into how the technology is actually used. The gap between the company’s promises and the audit logs remains wide.

Short searches. Long retention. Easy access. Minimal friction. The combination creates conditions where “LMAO” can stand in for legitimate police work. Until the reasons improve, the system will continue to invite skepticism. And more headlines like this one.

from WebProNews https://ift.tt/tZDNCSR