The Linux kernel community has agreed to eliminate the requirement that contributors explicitly label patches created with artificial intelligence tools.

The change emerged from a discussion among maintainers about the practicality of tracking AI-assisted contributions. Sasha Levin, a Linux kernel developer affiliated with Nvidia, announced that the project will no longer mandate the "Assisted-by" tag in commit messages, stating that the documentation requiring such disclosures is being deprecated.

The policy had previously instructed developers using any advanced coding tool to indicate its use through a specific tag. This rule was formally introduced in January 2026, following earlier discussions that began in July 2025. Its development coincided with broader industry conversations about AI's role in software engineering, including statements from Linus Torvalds, who had expressed cautious openness to AI tools while emphasizing the importance of code maintainability.

Greg Kroah-Hartman, a long-time Linux maintainer, noted at a recent summit that modern AI systems are now capable of producing code of sufficient quality for kernel development, particularly in identifying defects through pattern recognition. He observed that AI-generated code has reached a level of reliability that warrants its integration into established workflows, though concerns about error detection and accountability remain.

Despite this shift, some contributors continue to advocate for transparency measures. Laurent Pinchart, a kernel developer, argued that while the tagging system may be cumbersome, it serves a valuable purpose in setting clear expectations and maintaining accountability. He warned that removing the requirement without an alternative disclosure mechanism could weaken oversight, as some developers might otherwise avoid responsibility for AI-generated content.

Torvalds responded strongly to these concerns, dismissing the notion that AI tool usage should be restricted or singled out. He criticized the idea of imposing special rules for AI while allowing other development tools to operate without similar scrutiny. In a public message, he questioned the logic of such distinctions, suggesting that an overemphasis on tool provenance risks undermining practical development standards. He emphasized that understanding code is more important than the tools used to produce it, and that experienced contributors should be trusted to evaluate contributions on their merits rather than their origins.

The debate reflects broader tensions in the open-source community as AI tools become more embedded in software creation. While some view AI as a natural evolution of development practices, others remain cautious about its impact on code quality, security, and the long-term sustainability of collaborative projects. The removal of the disclosure rule marks a significant moment in this evolving landscape, signaling a pragmatic approach to integrating new technologies without compromising established norms.

Major banks including JPMorgan, Morgan Stanley, SMBC and MUFG are quietly attempting to transfer portions of the AI-related debt they originated during the recent data center construction boom, according to reporting by the Financial Times. The move signals that lenders who initially embraced AI financing are now confronting internal exposure limits and seeking to reduce risk through partial sales to external investors.

Six months ago, financial institutions competed aggressively to underwrite loans for data center projects tied to technology companies such as Oracle, Meta and others. At that time, these loans were seen as high-growth opportunities backed by major corporate tenants. Today, some of the same banks are exploring mechanisms to offload the riskiest tranches of these obligations while retaining only minimal exposure on their balance sheets.

One notable example involves a consortium that includes JPMorgan and MUFG, which has spent more than six months working to distribute approximately $38 billion in construction financing for a data center complex in Texas and Wisconsin leased to Oracle. According to Bloomberg, certain lenders are now considering selling portions of this debt to non-bank investors at a discount to expedite the process. The prolonged effort reflects growing discomfort with holding large concentrations of AI-related credit.

Matthew Moniot, co-head of credit risk sharing at Man Group, told the Financial Times that the scale of these transactions represents a level of exposure that current risk models were not designed to accommodate. His comment underscores the structural challenge faced by financial institutions as AI-driven infrastructure spending expands beyond initial expectations.

Oracle’s financial profile has shown signs of strain amid its aggressive expansion. Its five-year credit default swap spread rose to approximately 151 basis points in early October, a level not observed since 2009, according to market data cited by Bloomberg. In July, S&P downgraded the company’s credit rating to BBB-, just above speculative grade. The company has indicated plans to raise over $40 billion in debt and equity during the current fiscal year, building on the $18 billion in bond issuance completed in a single day the previous September. These funds are intended to support ongoing developments in Texas, Wisconsin and New Mexico.

Investor appetite for corporate debt tied to AI firms appears to be diminishing. Bonds issued by Alphabet, Amazon, Meta, Microsoft and Oracle now trade at a premium compared to similar-rated non-AI corporate debt, but order books for new offerings have contracted significantly. In February, demand reached about five times the amount of securities offered; by July, that ratio had fallen below two times, according to Bloomberg. This shift suggests a move from automatic demand to a more selective investor base.

Macroeconomic factors are compounding the environment. The 10-year U.S. Treasury yield reached 5.304% recently, its highest level since 2002, following a $39 billion auction that priced at 5.30%, the highest yield recorded since 2000. The Federal Reserve’s September meeting minutes referenced AI-related corporate borrowing as one of several contributors to rising rates, though analysts at Capital Economics attribute the broader trend primarily to inflation and fiscal deficits rather than technology sector financing.

The mechanics of debt distribution matter more than headline figures. Banks typically retain debt they believe will generate stable returns; the current activity reflects concerns about concentration risk. The original financing structure relied on special purpose vehicles to isolate risk, but that strategy has not shielded originating banks from growing exposure as AI infrastructure spending accelerates.

Investor preferences are becoming more discerning. Capital is gravitating toward data center projects leased by financially stable hyperscalers such as Microsoft and Google, while avoiding lower-rated or less-established tenants. This dynamic is reflected in pricing disparities: bonds backed by CoreWeave trade at yields between 7.00% and 9.875%, compared to 5.70% to 7.875% for those backed by hyperscalers, according to CCIR data. Smaller AI infrastructure firms seeking capital now compete with larger, more creditworthy issuers for limited investor interest.

The situation highlights a broader uncertainty about the sustainability of current financing models. While the AI buildout continues, the ability to secure favorable debt terms is no longer guaranteed. Six months of difficulty in offloading $38 billion of loan exposure provides tangible evidence that assumptions about open credit markets may be under real-time pressure, rather than remaining purely speculative.

Also read: Trump-linked stablecoin USD1 adds $427 million in a single week • IREN shares sink 20% in ten days as a scathing infrastructure report collides with Wall Street's AI bull case • Micron's Board Authorizes a Record $35 Billion Stock Buyback

This article is posted in Financial Markets News, check it out for more related stories.

A recent Hacker News discussion has reignited debate over passkeys, the password‑less authentication method gaining traction across the web. The controversy stems from a post titled "Passkeys are a trap. Do not use them," which argues that the technology could be co‑opted to enforce restrictive attestation requirements. While the post has been widely shared, experts say its warnings do not reflect how passkeys currently operate.

The core claim in the post is that passkeys support attestation — a cryptographic proof that a website can request about the originating device. The author frames this as a "hidden dial" that could eventually force users to rely only on approved software and operating systems, eliminating alternatives such as custom firmware or open‑source platforms. The argument also invokes unrelated topics, including uBlock Origin, the FIDO Alliance membership list and UK image‑scanning proposals, to suggest that passkeys are inherently anti‑user.

In practice, most consumer passkeys — those synced through iCloud Keychain, Google Password Manager, 1Password, Bitwarden or similar services — do not transmit any attestation data to websites. Because these passkeys are stored locally on the user’s device, a site that attempts to require attestation would instantly lock out the vast majority of its own users, an outcome that would be counterproductive for any service provider. Industry guidance for consumer sites has therefore discouraged the use of attestation for everyday logins.

When attestation is present, WebAuthn records the make and model of the authenticator via an AAGUID and a vendor‑issued certificate. The information is limited to the hardware’s identity; it does not reveal the operating system, browser, extensions, firmware version or any other environmental detail. Consequently, the notion that attestation can enforce a blanket policy about a user’s software stack is inaccurate.

The post’s recommendation to switch to hardware keys such as YubiKey is ironic, because those very keys are the primary source of attestation. YubiKeys ship with vendor‑issued certificates that enable exactly the kind of attestation the author claims to fear. While YubiKeys remain a solid choice for high‑security scenarios, their use does not avoid attestation — it embraces it.

The article also critiques the suggestion to revert to Time‑Based One‑Time Passwords (TOTP). Although TOTP is more secure than static passwords, it remains vulnerable to phishing attacks. Recent incidents, such as those involving the Evilginx phishing framework, demonstrate how attackers can capture both a password and its accompanying TOTP code and replay them in real time. Passkeys mitigate this risk because they are bound to the specific domain for which they were created; a passkey generated for example.com cannot be reused on examp1e.com, eliminating the possibility of domain‑confusion attacks.

Given these technical realities, security researchers advise continued adoption of passkeys where available. Users should store them in trusted password or passkey managers and remain vigilant about emerging policy debates around attestation. While pushing back against potential misuse of attestation is important, abandoning passkeys today offers no protective benefit and only exposes accounts to greater phishing risk.

The discussion underscores a broader need for clear, evidence‑based communication about emerging authentication technologies. As passkeys become more ubiquitous, distinguishing between genuine security concerns and speculative threats will be essential for users, developers and policymakers alike.

Apple has introduced a new Apple TV 4K model, the fourth generation of the streaming device, now available for pre-order at $299. The update centers on hardware improvements and expanded artificial intelligence functionality. The device features an A19 chip, which Apple states delivers up to 50 percent faster central processing unit performance and up to two times faster graphics processing unit performance compared to the prior model. This chip, previously used in the iPhone 17 series, includes a dedicated AV1 decoder intended to improve streaming efficiency and reduce bandwidth consumption.

The new model also introduces enhanced Siri capabilities through an integrated artificial intelligence system. Apple indicates that this AI supports contextual understanding, broad knowledge access, and expanded systemwide app interactions. Users can issue voice commands such as querying about Emmy Award-winning Apple TV content, identifying filming locations, evaluating content appropriateness for teenagers, or playing recommendations from specific shows. The fourth-generation device includes a three-microphone array with voice isolation technology, enabling hands-free interaction without pressing the Siri Remote button. This functionality is also being extended to second- and third-generation Apple TV 4K models through software updates.

The updated Siri Remote adds Precision Finding integration with the Find My network and includes a built-in speaker for audible location tracking. The remote is sold separately for $59 and remains compatible with earlier Apple TV 4K models. Additionally, the new Apple TV 4K supports a four-speaker surround sound system when paired with HomePod mini 2 or HomePod 2 devices, allowing automatic audio calibration based on room configuration and flexible pairing options across different speaker models.

Storage capacity has been standardized at 128 gigabytes, eliminating the previous 64-gigabyte base model. Connectivity upgrades include Wi-Fi 7 support, Bluetooth 6, Thread protocol compatibility, and standard Gigabit Ethernet across all models—features previously limited to higher-storage variants. The device runs on updated tvOS software, though specific version details have not been disclosed.

Pre-orders for the new Apple TV 4K are currently open. The company has not indicated a release date beyond the pre-order period, and further software feature rollouts, including potential enhancements to Smart Downloads in future operating system updates, remain under development.

A growing wave of public opposition to data centre construction is reshaping campaign strategies ahead of the US midterm elections.

Senator Chris Murphy, a Democrat from Connecticut, now convenes regular meetings with staff to assess the impact of data centres and draft legislation aimed at limiting their rapid expansion, according to a senior policy adviser who spoke on condition of anonymity. The shift reflects mounting constituent concern over the facilities' size, energy demands and community effects.

Residents in Indiana, Arizona, Michigan, Pennsylvania and Oregon have organized locally against proposed data centre sites, citing worries about land use, noise, water consumption and pressure on power grids. In Indiana, Angelita Soriano described feeling like "the guinea pig for this" after an Amazon data centre complex, comprising 26 buildings and 782 diesel generators, began construction across from her home. In Ohio, a consortium including OpenAI, Nvidia, SB Energy and the US Department of Energy plans a six‑square‑mile complex, roughly the size of a typical township.

Local zoning authority governs data centre approvals, leaving federal legislators with limited direct control. Nevertheless, opposition has intensified swiftly, prompting officials across the political spectrum to adopt clear positions. AI policy adviser Chloe Autio noted that resistance transcends partisan lines, with both parties feeling the pressure to respond.

Former President Donald Trump has dismissed the backlash as a "sick conspiracy," labeling opponents "backwards and poor." His stance has inadvertently bolstered Democratic candidates seeking to capitalize on voter discontent. Recent polling indicates that concerns about AI regulation and data centres rank among the top issues for voters, trailing only the Iran conflict and broader economic anxieties.

Industry proponents argue that data centres generate high‑quality jobs, yet research from the Brookings Institution finds that a new facility typically creates only 100–200 local positions, fewer than advertised, and exerts negligible influence on regional wages. Arizona Governor Katie Hobbs initially supported tax incentives for data centre projects but imposed a three‑year moratorium on such breaks after sustained public pressure. Similarly, Michigan Democratic Senate candidate Abdul El Sayed has integrated stricter data centre regulations into his campaign platform.

In Georgia’s gubernatorial race, a Democratic candidate advocates a pause on data centre development, while the Republican contender opposes any restrictions. Ohio Senator John Husted, facing a competitive re‑election battle against Democrat Sherrod Brown, has moderated his stance, emphasizing local decision‑making. National Republican Senatorial Committee memos now urge party members to reframe their messaging on data centres, describing the issue as a "sleeper" concern for the entire election cycle.

Texas illustrates the most pronounced reversal. Governor Greg Abbott once praised the state as the "epicentre of AI development" owing to its nearly 300 operational data centres and 170 pending projects. Within months, he cited "safety and quality of life" concerns and began reconsidering support for further expansion. Democrat James Talarico’s Senate campaign now calls for an end to "sweetheart tax deals" with Big Tech, positioning him as a frontrunner in a traditionally Republican race against Attorney General Ken Paxton, who maintains that ample space remains for new facilities.

Community resistance persists even when developers highlight sustainability measures, such as rain‑water capture and on‑site renewable energy. Balaji Tammabattula, chief executive of Barupon, noted that opposition can be immediate and visceral, with some residents telling him to "get the hell out of my state."

Moratoria and regulatory pauses are typically short‑term, leaving the long‑term trajectory of data centre growth uncertain. For now, the intensity of public sentiment suggests that data centre policy may influence voter behavior and campaign strategies in several key congressional contests.

Flock founder and CEO Garrett Langley maintains that the benefits of deploying surveillance technology at scale outweigh its risks, even as the company faces widespread backlash, contract cancellations and layoffs. Speaking on stage at TechCrunch Disrupt 2026 in San Francisco, Langley argued that a society with higher accountability for criminal behavior requires technological advancement, not pause. He described Flock’s license‑plate cameras and other recording devices as tools that help solve crimes, citing anecdotes from police chiefs and city leaders rather than addressing specific concerns about privacy or misuse. Langley compared the current controversy to the debate over artificial intelligence, suggesting both technologies are inevitable and that progress should not be halted by isolated incidents. He emphasized that Flock is not perfect and that mistakes have been made, but insisted the company’s impact on public safety justifies continued investment. When questioned about potential abuse, Langley offered no concrete safeguards, stating that preventing misuse should not disrupt law‑enforcement operations or legal processes. He also noted that some customers terminated contracts after believing Flock devices included facial‑recognition capabilities, a feature the company does not possess. Langley positioned Flock as a solution to staffing shortages in policing, particularly in cities like San Francisco where open positions remain unfilled. He concluded that, despite criticism, the technology improves lives for millions and that any trade‑offs in privacy are an acceptable cost for greater safety.

A single post on social media can quickly evolve into a widespread controversy, prompting organizations to seek tools that identify risks before they escalate. Gudea, a startup that describes itself as a "storm tracker for the internet," employs artificial intelligence to help companies monitor and understand how narratives spread online.

Unlike conventional social listening platforms that focus on metrics such as mentions, hashtags, and impressions, Gudea’s system maps the actors driving conversations, the pathways through which information travels, and the speed at which stories gain traction. Its AI analyzes behavior and relationships across more than 500 mainstream and fringe sites, including X, Reddit, and TikTok, to provide context beyond raw volume.

At TechCrunch Disrupt 2026, Gudea presented its approach on the Startup Battlefield stage, arguing that businesses need proactive methods to protect their reputations. CEO and founder Keith Presley noted that many communications leaders rely on legacy tools that only count activity without explaining its significance.

Presley, who previously worked on Andrew Yang’s presidential campaign and in defense-related projects, leads a team that includes Jonathan Sperber, chief strategy officer with a background in disinformation research, and James Thomin, chief technology officer who holds a Ph.D. in chemical engineering and has spent two decades in AI and machine learning. Presley described the venture as atypical for him, given his history of building political infrastructure rather than commercial enterprises.

The company originated from the George Washington Initiative, a nonprofit formed after the 2020 election that adapted intelligence strategies used abroad to counter online networks linked to extremist groups. Its founders identified a gap in existing tools that failed to reveal how information actually moved across people and platforms, leading to the creation of Gudea.

Three years after its founding, Gudea serves nearly 40 clients, including a Fortune 50 telecommunications firm, a Fortune 50 retailer, a major AI research laboratory, a prominent music group, public relations agencies, crisis communications firms, and government entities. Presley said the need for such technology has intensified with the rise of generative AI, which has lowered the cost and effort required to produce and coordinate inauthentic content. He observed that while early assessments found inauthentic activity in about one of four examined narratives, recent analysis shows that roughly nine of ten now involve coordinated manipulation.

Looking ahead, Gudea plans to introduce geolocation capabilities that can pinpoint where narratives emerge, down to the city-block level. It also aims to let customers test advertising campaigns through its AI model before launch, helping them anticipate potential negative outcomes. In a recent funding round, the company raised $7 million, bringing total investment to about $15 million, with Mudita Venture Partners and Silicon Road Ventures co‑leading the effort.

The organization emphasizes that its tools are designed to give decision‑makers clearer insight into emerging risks, enabling more informed responses without dictating specific actions.

A Denver-based startup, Intero Biosystems, is developing human intestinal organoids as a more accurate alternative to animal testing in drug development. The company creates lab-grown tissue that replicates the complexity of the human gut, aiming to improve how pharmaceutical and biotech firms evaluate drug safety and efficacy before human trials.

Unlike simpler models that mimic only fragments of the intestine, Intero’s organoids incorporate multiple cell types found in the human digestive tract, offering a more biologically relevant system for testing compounds. The flagship product, GastroScreen, is sold to pharmaceutical, biotech, food, and chemical companies for use in assessing how potential drugs interact with human intestinal tissue.

Each organoid measures about one millimeter in diameter and is shipped in vials containing roughly 100 units. Customers can use them in a healthy state or induce disease-like conditions to study responses to various substances. Intero also provides the growth media, reagents, and experimental protocols necessary for use, and offers custom study services, including experiment design and data analysis focused on toxicity and disease modeling.

Founded in 2024 and spun out of the University of Michigan, Intero has seven employees and serves fewer than 10 customers. The company reported generating over $400,000 in revenue from Q4 2023 through Q3 2024, following a $2.3 million pre-seed funding round.

Interest in non-animal testing methods has grown amid regulatory shifts. In April 2023, the U.S. Food and Drug Administration (FDA) announced plans to reduce reliance on animal models in drug development, highlighting alternatives such as organoids, organs-on-chips, and computational models. In March 2024, the agency released draft guidance on validating new approach methodologies (NAMs), establishing a framework for assessing their use in regulatory submissions without endorsing any single method.

Intero’s co-founder and CTO, Madeline Eiken, noted that while most drugs fail in clinical trials, current models—including animal studies and flat cell cultures—do not reliably predict human outcomes. The company believes its organoid platform better reflects human biology, potentially improving early-stage screening.

Other companies in the space include Altis Biosystems, which develops stem cell-derived intestinal models, and InSphero, which offers patient-derived organoids for gastrointestinal toxicity testing. MIMETAS uses organ-on-chip technology with microfluidic systems to simulate tissue environments. Intero differentiates itself through the complexity of its models, which are derived from induced pluripotent stem cells (iPSCs) rather than tissue biopsies. This allows for greater scalability and potential for personalization, as iPSCs can be generated from blood samples and differentiated into various gut cell types.

Intero is currently in the second phase of a three-stage validation process to demonstrate that GastroScreen can predict human drug responses more accurately than existing models. Early results show expected reactions to anti-inflammatory drugs and known gut-damaging compounds. The final stage will involve direct comparison with pharmaceutical companies’ animal and clinical data, using both established and novel drug candidates.

The company does not use artificial intelligence to build its organoids but plans to license the biological data generated from them for use in AI-driven drug discovery. Short-term goals include scaling production from thousands to hundreds of thousands of organoids per batch through automation. Long-term ambitions include personalized medicine applications, where a patient’s own cells could be used to test multiple drug responses prior to treatment, and the eventual development of a fully synthetic intestine.

Intero presented its technology at TechCrunch Disrupt, where it competed for the Startup Battlefield title. The company’s progress reflects a broader industry shift toward human-relevant models that could reduce animal use and improve the translatability of preclinical research.

Apple introduced HomeView, a $299 smart display designed as a fixed, always-on household interface. The device resembles a larger Apple Watch mounted on a metallic base similar to the HomePod Mini. Unlike tablets used for smart home control, HomeView operates without a battery, remains permanently powered, and stays in a designated location, enabling consistent access for multiple household members.

Initial reactions questioned the need for a dedicated device when iPhone and iPad functions could replicate many features. However, HomeView addresses limitations of personal devices, such as battery dependency, lack of multiuser support, and the need for constant login to iCloud accounts. It serves as a dedicated hub for Apple Home, offering always-available access to calendar, music, and smart home controls without requiring users to retrieve personal devices.

The device’s fixed placement allows for straightforward interaction with home automation systems, such as adjusting thermostats or checking family schedules. Its always-on nature eliminates the need for users to locate or power devices, reducing friction in daily use. Apple emphasizes multiuser functionality, enabling shared access without compromising individual privacy settings.

Siri AI powers HomeView’s conversational interface, designed to perform tasks without excessive follow-up prompts. According to Lilian Rincon, Apple’s vice president of product marketing for AI, the system prioritizes direct action over interrogative responses, aligning with user expectations for natural interaction. This approach distinguishes it from competitors who frequently request additional information.

HomeView also integrates video calling capabilities, allowing household members to receive calls even when personal devices are unavailable or silenced. This feature addresses common communication gaps in family settings, such as missed calls during busy periods.

Customizable display faces and live security camera feeds enhance its utility as a central home interface. While early reviews noted limited advanced automation features, Apple positions HomeView as a simplicity-focused solution for households prioritizing ease of use over technical complexity. The device targets families already invested in Apple’s ecosystem, aiming to streamline shared digital tasks without introducing new interaction paradigms.

By focusing on accessibility and consistent availability, HomeView represents Apple’s strategy to expand its presence in the smart home market through a device optimized for communal use rather than individual consumption.

Flock chief executive Garrett Langley told TechCrunch that the company’s audio sensors are designed to capture only gunshots, a claim that contrasts with the company’s public description of broader detection abilities. At TechCrunch Disrupt, Langley explained that the Gunshot Detection system was created to address a gap in emergency response: most people do not call 911 when they hear a gunshot, so the product was built to provide earlier alerts for public safety incidents. He added that a small group of customers later requested expanded alerts for events such as abductions or assaults, prompting the company to prototype additional features. Langley said the prototype was limited to gunfire detection and that the system was never intended to monitor conversations or analyze speech.

Flock’s website, however, lists capabilities that extend beyond gunfire detection. The site states that the audio systems can "detect street takeovers early to stop chaos before it escalates," "identify severe crashes in real time so help arrives faster," and "expand awareness beyond gunfire with alerts for car crashes, fireworks, and sideshows." An FAQ section further notes that the detection library can include fireworks, sideshows, and other disruptive community events, and that the library is "continually expanding to include emerging threats relevant to customers’ specific needs."

When asked about the discrepancy between his statement that the product only detects gunshots and the website’s broader claims, Langley responded, "It’s just gunshots." He maintained that the product’s core function is limited to gunfire detection, while acknowledging that some customers had requested additional alert types. A Flock spokesperson, Paris Lewbel, told TechCrunch that the technology is "primarily used to detect gunfire but is also capable of detecting fireworks, sideshows, and potential car crashes," and that Langley’s remarks were meant to emphasize that the system does not monitor conversations or analyze speech and does not currently detect human screaming or distress.

The company’s audio collection has faced scrutiny from privacy advocates and the public. Critics, including the ACLU, have raised concerns about mass surveillance, leading to protests outside the TechCrunch Disrupt venue where Langley spoke. Some jurisdictions, such as Texas and Florida, have moved to end contracts with Flock, and Amazon’s Ring division has terminated its partnership with the company. As of 2024, Flock reported having a number of microphones comparable to ShotSpotter, which had approximately 25,000 microphones in operation, according to reporting by Wired.

Langley said the company is responding to customer feedback and adjusting its approach, stating that it has removed certain claims from its marketing materials after discussions with users who expressed concerns about the scope of detection.

OpenAI is facing lawsuits from eight U.S. states that accuse the company of endangering minors and misleading consumers about the safety of its AI chatbot, ChatGPT. The states — Iowa, Kansas, Nebraska, West Virginia, Arkansas, Arizona, Idaho and Montana — filed separate complaints alleging violations of local consumer‑protection statutes. Each lawsuit claims OpenAI failed to warn users about potential hazards of the technology and continues to market it as a safe, youth‑appropriate tool while allegedly exposing vulnerable users to harm.

The complaints cite several incidents in which families of teenagers and adults have reported that the chatbot contributed to fatal outcomes. State attorneys general contend that OpenAI’s representations of safety were false, describing the system as "dangerous to vulnerable users" because it can degrade critical thinking, encourage paranoid or delusional thinking, provide unsafe medical advice, and promote unhealthy behaviors and relationships. In addition, the states allege that OpenAI collected personal data from minors without adequate safeguards and allowed underage users to access the service.

Recent high‑profile AI‑related security breaches are referenced as evidence of what the plaintiffs describe as OpenAI’s "recklessness." Arizona’s filing states that OpenAI has shifted from its original nonprofit mission of safety‑first development to a profit‑driven model that neglects essential safeguards. The lawsuits seek civil penalties, an injunction against deceptive marketing of ChatGPT’s safety, and the removal of any personal data unlawfully gathered from children.

In a statement to The Verge, OpenAI spokesperson Drew Pusateri said the company has "built industry‑leading age‑specific protections" into ChatGPT, including parental controls, break reminders and model safeguards designed for younger audiences. While acknowledging that no system is perfect, the company said it will continue to work with experts to improve safety and collaborate with states on broader AI policy that applies across the industry rather than targeting a single firm.

The litigation remains ongoing, and the states have not yet disclosed the specific monetary penalties they will pursue.

A YouTuber known for satirizing AI-driven hiring tools recently demonstrated how such systems can malfunction when handling real-world interactions. In a video, the creator, who uses the pseudonym Graham Dot Zip, conducted an interview using an AI agent he developed to represent himself. The AI incorrectly stated his professional background and fabricated salary expectations that conflicted with his actual requirements. When the AI refused to adjust its position, the conversation stalled, with both systems repeating polite phrases and failing to conclude the interaction. The incident highlights concerns about AI's capacity to manage nuanced negotiations and accurately convey human qualifications. While the scenario was framed humorously, it underscores broader questions about the reliability of AI in professional contexts, particularly as companies explore deploying such tools for recruitment and commercial transactions. The video contributes to ongoing discussions about the limitations of current AI systems in handling tasks requiring contextual understanding and ethical judgment.

Ukrainian drone strikes targeted three Yandex data centers in Russia between October 7 and 11, causing near-total network collapse for the company's search and cloud services. Cloudflare analysis shows traffic from Yandex Search (AS13238) fell 99% below baseline on October 11, while Yandex.Cloud (AS200350) dropped 90% during the same period. Both networks remained at 85-90% below normal levels through October 13, with partial recovery observed by October 14.

The impact varied significantly between the two autonomous systems. Yandex Search traffic consisted almost entirely of verified search crawlers indexing web content, with 78.1% of requests identified as YandexBot. This led to a 44% decline in traffic to publishing sites, where 90% of pre-event requests were crawler activity. In contrast, Yandex.Cloud handled diverse automation, with 95% of requests classified as bots but none verified. Its traffic decline reflected shifts in attack patterns rather than cessation of malicious activity.

Analysis of request behavior revealed key insights. While vulnerability scanning and plugin probing dropped sharply during the outage, similar activities reappeared from alternative Russian hosting providers like Radist and Selectel within hours. The displaced automation maintained identical technical signatures, including shared JA4 TLS fingerprints and custom User-Agent strings, indicating coordinated migration rather than disruption.

The data contradicts claims that the strikes crippled Russian bot operations or social media manipulation efforts. Global bot traffic from Russian IP addresses declined proportionally to Yandex's offline capacity, but no evidence showed a sustained reduction in malicious activity. Instead, automated workloads simply relocated to unaffected infrastructure, preserving overall request volumes and attack patterns.

This event underscores limitations in relying on traffic volume metrics to assess cyber threats. Disruptions to physical infrastructure do not equate to dismantling automated threat actors, who demonstrate resilience through rapid migration capabilities. Effective defense requires behavioral analysis of request patterns rather than monitoring aggregate traffic from specific network providers.

Bungie’s head of project, Josh Deane, recently shared a clip showing Destiny 1 running at 4K, 60 frames per second and a 105‑degree field of view on PC. In a quote‑tweet, Deane wrote, “Seems like a pretty game, wonder if it will do well…” The timing coincided with Bungie’s announcement to unvault Destiny 2 content and its confirmation that new Destiny titles are in development. Long‑time franchise reporter Paul Tassi observed that the clip represents the clearest indication yet that Bungie may be exploring a revival of the original game. The speculation follows a surge of veteran players returning to Destiny 1 while Destiny 2 faced challenges after its recent expansions. Although no official announcement has been made, the tease aligns with broader efforts to restore legacy content and expand the franchise’s long‑term support. A potential remaster would likely incorporate expansions such as The Taken King and Rise of Iron, along with quality‑of‑life improvements requested by the community. KitGuru notes that a decade‑long wish for a PC version of Destiny 1 may be moving closer to reality, though the current information remains preliminary.

A recent Government Accountability Office report suggested that Immigration and Customs Enforcement officers should have their body-worn cameras turn on automatically whenever they draw a firearm. The recommendation came after a series of ICE shootings drew public scrutiny. While the Department of Homeland Security accepted most findings in the GAO analysis, it rejected the camera activation proposal. DHS financial management director Jane Doe cited concerns that automatic activation could create operational hazards, such as unintended recordings during normal patrols, and might capture sensitive interactions with informants or victims without consent. The agency did not elaborate on alternative oversight measures it plans to adopt.

Bulgarian Prime Minister Rumen Radev confirmed on Wednesday that a fire had erupted at a Starlink satellite ground station located near Sofia last week. Two individuals of German citizenship were taken into custody in connection with the incident, according to reports by Bulgarian media citing investigators.

The Starlink network serves as a critical communications infrastructure for Ukraine’s military operations amid its ongoing conflict with Russia. The facility affected by the blaze was privately owned, officials noted, and housed 17 antennas, 11 of which were deliberately ignited on October 9, as reported by the pro-Western opposition alliance Democratic Bulgaria (DB).

Bulgarian authorities indicated that the suspects, both aged 20 and 21, entered the country legally through the Kulata border crossing after traveling from Greece. Their backgrounds were described as migrant, though no further details were released. Investigators are examining whether the pair are connected to a broader sabotage network, potentially linked to two other men detained in Germany following an earlier attack on a Starlink terminal near Warsaw.

A separate fire had previously occurred at a Starlink ground station south of Warsaw on September 23. Polish authorities suspect Russian intelligence services orchestrated that incident, and two suspects were arrested in Nürtingen, Germany, earlier this week. Bulgarian Interior Minister Ivan Demerdzhiev informed parliament that heightened security protocols have been implemented at similar facilities, though responsibility for their protection remains with private owners.

Plamen Tonchev, head of Bulgaria’s State Agency for National Security (DANS), stated that sabotage cannot be ruled out as a recurring threat. He emphasized that such actions pose risks to critical infrastructure without confirming any direct state involvement. The investigation remains ongoing, with no official attribution released to date.

The incident underscores growing concerns over the vulnerability of satellite communication assets used in conflict zones, particularly those supporting Ukraine’s defense efforts. While no group has claimed responsibility, the pattern of attacks across Central and Eastern Europe reflects heightened geopolitical tensions affecting critical technology infrastructure.

Rockstar Games now employs 4,711 people across 12 offices worldwide, a figure that represents nearly a doubling of its staff since the 2018 launch of Red Dead Redemption II. The company’s growth accelerated over the past eight years as it assembled a team for Grand Theft Auto VI and reorganized its development pipeline. A Bloomberg feature confirms that thousands of new hires were added during GTA VI’s creation, including a Los Angeles studio dedicated to NPC writing, animation and production. Additional expansion has been noted in India and other regions as the game’s systems grew more intricate. The studio’s internal policies have attracted criticism, particularly the elimination of remote work and limits on accessing development tools outside company premises. These moves coincided with efforts to tighten scheduling, curb overtime and respond to concerns that surfaced after Red Dead Redemption II. Executives have argued that the larger workforce was required to keep GTA VI on schedule. Nevertheless, heightened security measures did not prevent leaks; a working build of the game has been circulated online, with gameplay videos appearing for months. The ambitious scope of GTA VI, featuring dual‑protagonist AI and a high‑fidelity world simulation, demanded extensive resources, and Rockstar says the expanded team was essential to meeting those targets, with nearly 5,000 staff contributing across its global offices. KitGuru notes that the rapid growth helps explain why GTA VI has become one of the most costly and labor‑intensive games ever produced.

A Sydney-based startup called Apate has built a system that uses thousands of AI agents to pose as potential victims and engage scammers in real-time conversations. The agents answer phone calls, reply to messages, follow phishing links, and join social media groups while adopting varied personas across age, gender, and language. Their goal is to keep fraudsters occupied long enough to extract intelligence such as bank account details, cryptocurrency wallet addresses, phone numbers, and impersonated brand information.

Apate’s platform has reportedly consumed 17 years of scammers’ time across 2.5 million interactions. The company serves clients including banks, telecom providers, government agencies, and utilities, who use the gathered intelligence to strengthen their fraud prevention strategies. According to co-founder Brad Joffe, most fraud defenses operate reactively—blocking numbers or monitoring transactions after harm has occurred. Apate instead engages scammers directly to collect actionable data before victims are targeted.

The concept originated from a personal experience. Apate’s founder, Dali Kaafar, once pretended to be an unsuspecting victim during a scam call to amuse his children. The 44-minute exchange, which diverted the scammer from potential real victims, inspired him to explore automation. Research conducted at Macquarie University, where Kaafar previously worked, further developed the technical foundation using machine learning and natural language processing to analyze scam communications.

Apate employs both large language models and custom-trained models, fine-tuned on actual scam interactions and the university’s research. Its agents can be assigned specific objectives, such as extracting a crypto wallet address or identifying fraudulent websites. They also scan forums, investment groups, and social platforms to detect emerging scam activity and initiate contact under fabricated identities.

Clients include Commonwealth Bank of Australia and TPG Telecom. Joffe reported that TPG has redirected nearly 590,000 calls to Apate’s agents, while Commonwealth observed a 76% reduction in publicly disclosed scam losses. An unnamed Australian financial institution used outbound agents to disrupt 22 separate scam campaigns. The startup has raised nearly $10 million in total, including a $8.15 million seed round in August from investors such as Lobby Capital and OIF Ventures. Plans for expansion include establishing a London office and increasing U.S. operations.

The company’s name, drawn from the Greek deity of deception, reflects its operational premise. Joffe noted that scammers frequently express frustration through profanity directed at the AI agents, a behavior the team monitors as an indicator of engagement. Each instance of swearing is recorded as a metric of success within their internal analytics.

While AI has enabled more sophisticated scams—such as deepfake voice impersonations and automated social engineering—the same technology is now being leveraged to counteract them. The FBI reported over 22,000 complaints related to AI-facilitated crimes in 2025, resulting in nearly $893 million in losses, with investment scams accounting for the majority of financial harm. Apate’s approach represents a shift from passive defense to proactive intelligence gathering, using AI not to deceive consumers, but to expose and disrupt criminal networks.

Rockstar Games announced that the development work on Grand Theft Auto VI has produced a new technical foundation intended to accelerate the creation of subsequent titles. After an eight‑year development cycle, senior staff including Phil Hooker and Aaron Garbut told Bloomberg that the project involved a complete overhaul of the studio’s internal technology. They described the effort as a “foundational” rebuild that now provides a stable base for future content, with Hooker noting that upcoming projects can “go more sideways than upwards” because of the systems established during GTA VI’s development. The long production timeline included two internal delays, significant staffing changes and an expansion of Rockstar’s global workforce. The company also highlighted a shift in its release cadence: whereas it previously shipped major titles every two to four years and occasionally in consecutive years, the current structure of nearly 5,000 employees results in each new project requiring six to eight years to complete. This makes the promise of faster development a notable departure from past practices. Rockstar drew a parallel to the transition from Grand Theft Auto III to Vice City, which was possible in under a year because the underlying technology was already in place. With GTA VI now serving as the studio’s new baseline, the team believes future releases will not require another full rebuild of its systems. KitGuru observed that it is somewhat unexpected for Rockstar to have needed to develop much of its technology for GTA VI even after the technical achievements demonstrated by Red Dead Redemption 2.

Silicon Valley executives have begun describing smoking as a symbolic gesture of confidence in artificial intelligence, according to statements made during a CBS interview. Palantir chief technology officer Shyam Sankar told CBS News' Margaret Brennan that some AI researchers have taken up smoking to signal their belief that AI systems will eventually solve major health challenges, including cancer. Sankar described this behavior as a 'bet' that AI breakthroughs will arrive imminently, comparing it to placing 'chips on the table' to demonstrate commitment to the technology.

The remarks came during a discussion about effective altruism, a philosophical approach emphasizing long-term, data-driven decisions to maximize societal benefit. Sankar characterized certain interpretations of effective altruism as fostering 'millenarian doomsday cult' mindsets among AI developers, particularly those expressing extreme optimism about artificial general intelligence. He suggested that some proponents fill a 'god-shaped hole' in their worldview with faith in AI's eventual ability to 'cure cancer' or other ailments.

Sankar referred to the practice as 'mass psychosis' and 'social delusion,' arguing it distracts from substantive technical discourse about AI capabilities and limitations. He noted that only about 2,000 AI researchers operate in the United States, with most adhering to effective altruist principles that prioritize long-term existential risks over immediate practical concerns.

A cancer researcher who spoke with The Register expressed skepticism about the smoking symbolism, stating that developing new cancer treatments typically requires years of research and testing, regardless of AI advancements. The researcher emphasized that AI's role in drug discovery remains largely experimental, with no proven timeline for delivering breakthrough therapies. They also highlighted the health risks of smoking, noting that non-smoking associates and family members would inevitably face exposure to secondhand smoke.

The comments reflect a broader tension between technological optimism and practical risk assessment in the AI industry. While some leaders frame extreme beliefs as necessary bets on AI's future, experts caution that such narratives may obscure realistic timelines for medical innovation and overlook established health hazards associated with smoking.

Deezer announced plans to introduce a toggle that will allow users to filter out AI-generated music from their platform, with the feature slated for release before the end of the year. The company stated that enabling the toggle will prevent AI-generated tracks, albums, and related artist recommendations from appearing in search results. Deezer CEO Alexis Lanternier said the move responds to user demand for greater control over content, noting that the rapid growth of AI music on the service was a key driver for the feature. A 2023 survey cited by the company found that approximately 45% of users expressed interest in removing AI-generated content from their feeds. Deezer had previously implemented AI-detection tools and reported that AI-generated tracks accounted for 10% of daily uploads at the time. By July, the company reported that more than half of daily uploads — averaging 90,000 tracks — were AI-generated. Other platforms have taken varied approaches: Bandcamp has banned AI tracks entirely, Tidal has discontinued monetization for AI uploads, and Spotify has opted to label AI content more prominently rather than remove it. In August, Variety reported that Apple Music is exploring labeling mechanisms for AI tracks by year’s end. Industry analysts note that user-driven tools for identifying AI music are also gaining traction. Ivan, a technology reporter covering global consumer trends, is based in India and has contributed to outlets including Huffington Post and The Next Web.

Paul Henninger, KPMG's head of technology and data, informed staff of his promotion to group head of consulting in an email sent before the firm disclosed broader restructuring plans. The message arrived as approximately 200 employees in AI, cyber, SAP and testing roles faced redundancy, with the cuts affecting about 4 percent of the division.

Henninger outlined his new responsibilities, stating he would lead the Consulting practice and work with his leadership team to strengthen a proposition‑driven market approach, accelerate AI‑enabled capabilities, and develop specialized sales, technical and delivery career pathways. He also noted that he would hand over current duties and assist the team in preparing for the transition.

The email acknowledged the difficult timing for staff undergoing consultation, urging continued mutual support and use of available resources. An interim leader will oversee the Tech and Data unit during the transition, reporting directly to Henninger, while KPMG assured that reporting lines and business priorities would remain unchanged for the time being.

KPMG confirmed that 200 positions would be eliminated before the holiday season, and some affected employees criticized the severance offer of £100 in exchange for a waiver of legal claims as insufficient. A spokesperson for the firm described Henninger as a respected industry leader who has brought innovation and AI to the forefront of consultancy services, citing the company's 2 percent revenue increase to £3.6 billion in its UK and Swiss operations and a 14 percent rise in profit before tax to £576 million.

Jonathan Holt, group chief executive of KPMG in the UK and Switzerland, added that the firm continued to invest in its people, skills and talent pipeline, increasing colleague bonus spend by 18 percent across the Group and expanding learning opportunities. The spokesperson emphasized that elevating expertise in consulting was a natural response to growing client demand for innovation.

The situation reflects a complex interplay of leadership changes and workforce reductions within a major professional services firm operating in a competitive technology market.

A growing number of employers have begun omitting cover letters from job postings, and surveys reveal sharply divided views on whether hiring managers still read them. The shift accelerated as job seekers increasingly turn to artificial intelligence tools to draft application materials, according to Judd Kessler, a professor of business economics and public policy at the University of Pennsylvania.

"It’s not just easier to write a cover letter, it’s easier to write a phenomenal cover letter," Kessler said. "That’s become more of a prerequisite and less of a differentiator."

LinkedIn Research reported that for every open role projected in 2026, there will be twice as many applicants as there were for comparable positions in 2022. The surge in applications has made recruiters less likely to devote time to reviewing cover letters, said Johnny C. Taylor, president and chief executive officer of the Society for Human Resource Management.

"The cover letter is literally dead, never to be resuscitated," Taylor said. "There’s too much volume for this beautiful piece of prose."

Some organizations now employ AI systems to screen applications, creating a feedback loop in which automated tools evaluate cover letters that were themselves generated or edited by AI. Pelumi Ogunlana, a corporate banking professional in the Chicago area, described instances where résumés and cover letters displayed visible edits from chatbot platforms.

When a posting marks a cover letter as optional, candidates face uncertainty about whether to include one. If submitted, the document must stand out from the flood of AI-generated submissions, said Jason Baker, vice president of talent acquisition at Arctic Wolf, a cybersecurity firm. Baker cited an example of an applicant who did not secure the original role but was later reconsidered because their cover letter demonstrated a distinct human touch.

Vicki Salemi, a career expert at Monster, noted that the expectation for cover letters varies widely across industries and individual managers. While some hiring teams rely solely on résumés, others continue to scrutinize every cover letter before making decisions. Tailoring each letter to the specific position remains a recommended strategy, even if its impact cannot be guaranteed, she said.

Kessler anticipates that as AI lowers the cost of applying to jobs, employers will increasingly depend on alternative signals of candidate quality, such as personalized recommendation letters or participation in in-person networking events — activities that AI cannot replicate. "These are things an AI agent cannot do on your behalf," he said. "These are not things you can do for every firm."

OpenAI confirmed on Tuesday that a concealed recording device had been found inside one of its San Francisco offices, prompting an investigation and the arrest of an employee. The company stated it identified the device as sophisticated and employed internal tools to analyze its origin and determine how it was placed within the premises. Following the discovery, OpenAI reported the incident to the Federal Bureau of Investigation. The statement did not specify the nature of the employee's alleged involvement or provide further details about the investigation. This development adds to a series of privacy-related incidents involving the company, which has previously faced scrutiny after cyberattacks attributed to foreign actors targeted its systems. OpenAI emphasized its commitment to cooperating with law enforcement and maintaining the security of its infrastructure.

Apple announced an upgraded Health app in September, describing it as a major redesign that would add features such as health age calculations and the ability to schedule laboratory tests through Quest Diagnostics. The company has consistently said the revamp would arrive later this year.

The first version of the redesigned app appeared in iOS 27.2 developer beta 1, and was later included in both the public beta 1 and public beta 2 releases. Early testing showed the new interface incorporating additional data categories and a scheduling function for lab work.

Subsequent developer builds, including iOS 27.2 beta 4, restored an earlier version of the app that omitted the redesign. Release notes for that build indicated that health data entered during the beta period would remain accessible and could be viewed in the Settings app. Apple also reaffirmed that the full update remains slated for release later in the calendar year.

Analysts suggest the revision may be postponed to a later beta, possibly iOS 27.3, with a tentative window in late October or early November. The delay reflects Apple’s practice of iterating on complex health‑related features before broader deployment.

The update will retain any health data collected during the beta testing phase, allowing users to continue tracking without loss of information. No official launch date has been provided, and Apple has not indicated any changes to its overall timeline for the health‑app overhaul.

According to TrendForce's latest industry analysis, worldwide notebook shipments are expected to decline by a single-digit percentage in 2027. This projection coincides with a strategic shift among laptop makers who are evaluating production locations amid rising operational costs that are making manufacturing outside China less economically viable. TrendForce forecasts that the proportion of global notebook production occurring outside China will decrease from 24 percent in 2025 to 21 percent in 2026, and could dip below 20 percent by 2027. The primary challenge identified is a transition from concerns over component supply to a more pronounced weakening in consumer demand, as manufacturers continue to navigate elevated pricing for central processing units, memory modules, and solid-state drives. The research firm further noted that if component costs remain elevated, brands may increasingly transfer these expenses to end users.

The Federal Communications Commission has granted Netgear a conditional import approval for consumer routers, cable modems and gateways through October 1, 2027. The approval comes despite the company's manufacturing base remaining in Asia and no announced plans to shift production to the United States.

The FCC's conditional approval cites a "specific determination" by the Pentagon that these devices do not pose risks to U.S. national security. This marks a reversal from the agency's earlier justification that foreign routers inherently threatened security due to incidents like the Volt Typhoon cyber espionage campaign, which primarily targeted Netgear equipment.

The approval mechanism requires manufacturers to submit detailed plans for U.S. manufacturing expansion and capital investment timelines. Netgear has not publicly disclosed such plans, and neither the company nor the FCC responded to inquiries about whether these requirements were met. The FCC's conditional approval specifically covers Netgear's Nighthawk, Orbi and related router families, as well as cable gateways and modems.

In its SEC filing, Netgear suggested the approval would allow indefinite software updates to existing routers, but the FCC clarified that such updates do not require agency approval unless they alter radio performance. Netgear's CEO also characterized the decision as advancing "stronger safety and security standards," though the FCC's process does not evaluate safety or security criteria.

The situation raises questions about the relationship between import approvals and domestic manufacturing commitments, particularly as companies navigate evolving regulatory frameworks for networking hardware.

Update: Neither Netgear nor the FCC provided comment in response to requests for clarification about manufacturing plans and compliance with conditional approval requirements.

Asus has confirmed that the entry-level configuration of its ExpertBook A7 will ship with 12GB of RAM, correcting an earlier announcement that specified 8GB. The company maintains the $899 price tag for the model, which is scheduled to launch in January 2027. During a briefing, Asus representatives initially indicated the base model would include 8GB of memory, but later updated the specification to 12GB without revising the advertised price. The device also features an OLED display and a haptic trackpad. Industry observers note that the change reflects growing demand for higher memory capacity in premium ultrabooks, though the company has not provided a detailed rationale for the adjustment. Consumer response has been generally positive, with several tech reviewers highlighting the value of the upgraded specification at the same price point. Asus has not indicated plans to modify other configurations or pricing tiers at this time.

Newegg has reduced the price of Intel's Arc B580 graphics card to $329, positioning it as the lowest‑priced 12 GB GPU currently available. The retailer is offering the card as part of a limited‑time promotion that also bundles an ASRock Challenger 360 mm all‑in‑one liquid cooler, normally priced at $89. The bundle effectively lowers the total cost to approximately $240, which is below the card’s original launch price of $249. The Arc B580 features 20 Xe2 compute cores, 20 ray‑tracing cores, 12 GB of GDDR6 memory on a 192‑bit bus, and delivers up to 456 GB/s of bandwidth. In many modern titles it can exceed 60 FPS at 1080p resolution with ultra settings, and it supports Intel Frame Generation and Multi‑Frame Generation technologies. The discount comes amid broader market conditions where newer generation GPUs, such as the RTX 5050, are selling for $400 or more, making the B580 an attractive option for budget‑conscious builds targeting sub‑$1,000 systems.

Go 1.27 introduces significant improvements to HTTP/1 connection reuse, reducing latency in network operations. Testing revealed that calling a simple HTTP GET function 100 times against index.golang.org took approximately 710 milliseconds in Go 1.26 but only 150 milliseconds in Go 1.27 on the same hardware, representing nearly a fivefold speedup. This optimization stems from how Go handles unread response bodies during connection reuse. In HTTP/1, connections can only be reused if the entire response body is consumed before initiating a new request. Earlier versions of Go failed to properly manage partially read bodies, causing connections to be abandoned unnecessarily. For instance, when only the status code was needed—such as in health checks—developers often overlooked reading the full response body, leading to repeated connection overhead. Go 1.27 addresses this by automatically draining unread bodies asynchronously within a 50-millisecond window or up to 256 kilobytes, whichever comes first. This change eliminates the need for manual workarounds like custom draining functions, which previously introduced complexity and potential bugs. The net/http Transport now handles these edge cases internally, ensuring more efficient reuse of existing TCP connections. While HTTP/2 has long avoided this issue through multiplexed streams, HTTP/1 remains prevalent in many environments. The update maintains backward compatibility while enhancing performance for common use cases. Users are advised to continue closing response bodies as needed, trusting the standard library to manage connection reuse efficiently without additional intervention.

Cisco and Nvidia have unveiled a new optical pluggable module designed to deliver up to 12.8 terabits per second of bandwidth, targeting the growing demands of artificial intelligence data centers. The ultra-dense small form factor pluggable (USFP) standard, announced at the Open Compute Project global summit, aims to consolidate high-speed connectivity into a compact form factor roughly equivalent in size to existing optical modules. Each USFP module supports up to 32 lanes of 400 gigabits per second, collectively enabling 12.8 Tbps of Ethernet throughput. This capacity surpasses current high-end pluggables, which typically max out at around 1.6 Tbps using eight 200 Gbps links.

The collaboration seeks to streamline cabling infrastructure within AI clusters, where individual GPUs often require 800 Gbps or higher network interface connections. By consolidating connectivity for multiple GPUs into a single fiber bundle, the USFP module could reduce cable complexity in dense server environments. For instance, a future switch capable of 204.8 Tbps could support 128 endpoints using just 16 USFP ports, potentially simplifying network architecture.

Technical challenges accompany the performance gains. Cisco’s USFP specification permits up to 100 watts of power consumption per module, significantly higher than the approximately 20 watts typical of contemporary pluggables. The design employs a two-stage insertion mechanism to ensure reliable electrical contact across multiple high-speed lanes, though this increases mechanical complexity. Additionally, a single faulty component in the module could affect a larger portion of the network compared to traditional pluggables, though this risk is considered manageable relative to failures in co-packaged optics, which might disrupt entire server racks.

Nvidia’s involvement marks a notable shift, as the company has largely transitioned to co-packaged optics in its latest switch designs to improve energy efficiency. However, Nvidia continues to rely on pluggables for its network interface cards, citing flexibility for diverse deployment scenarios, such as top-of-rack switches using lower-power copper cabling. The company’s upcoming Vera Rubin Ultra platform, announced in early 2025, targets 115.2 terabits per second of inter-rack bandwidth, aligning with the 6.4 Tbps per GPU bandwidth offered by Cisco’s initial USFP specification.

The USFP standard is being released as an open specification, allowing broader industry adoption. Cisco has also emphasized compatibility with SONiC, a widely used open-source network operating system, extending its existing support for the platform. This includes plans to integrate its Live Protect security patching system into SONiC, with rollout expected in the following year. Meanwhile, Microsoft has indicated use of Cisco’s Silicon One P200 routing chips to enable multi-datacenter AI workloads, reflecting growing interest in distributed computing architectures that bypass local power limitations.

The development underscores a broader industry push to scale network bandwidth while managing power and physical constraints in AI infrastructure. As AI models grow in size and complexity, the demand for high-speed, low-latency interconnects continues to drive innovation in optical networking, with USFP representing one approach to balancing performance, efficiency, and deployability.

Framework's updated Laptop 12 enters a more crowded entry-level market than when it first launched in 2023, aiming at students and budget-conscious buyers who value repair and upgrade options. The new model retains the company's focus on modularity, allowing users to replace components like the motherboard and storage without full device turnover. However, its base price now starts near $1,000, and configurations often require additional purchases of RAM and SSD to reach usable performance levels. The 1080p60 display remains fixed and lacks premium features, while the webcam continues to deliver only average quality.

Senior CPU Analyst Jake Roach of Tom's Hardware noted that while upgradability remains a core strength, limitations persist. The system uses a single-channel memory interface, capping bandwidth regardless of RAM capacity, and the display cannot be upgraded. These constraints place it behind some competitors in raw performance and flexibility, though Framework devices like the Laptop 13 still offer dual-channel memory and multiple display options.

User discussions highlighted broader concerns about sustainability and long-term value. One commenter noted they had upgraded the motherboard on their Laptop 13 twice, extending its life and reducing electronic waste. Others questioned why repair metrics like iFixit scores were not included in the review for comparative context. While some critics argued the device is less repairable than Apple or Dell systems, Framework maintains that its design enables user-led repairs with publicly available manuals and standardized parts.

The Laptop 12’s positioning reflects a niche strategy: it does not compete directly on price or specs alone, but on repairability and adaptability in an era of rising device costs and environmental awareness. Its success will depend on whether users prioritize longevity and customization over immediate performance or aesthetic refinements.

Suicide rates among middle-aged men have reached their highest recorded levels in recent years, according to data released by public health authorities. The increase follows a decade-long trend that has seen deaths of despair rise across multiple age groups, with men between 35 and 54 showing the most significant growth. Researchers note that economic pressures, social isolation, and limited access to mental health resources have contributed to the pattern, though causality remains complex. The Centers for Disease Control and Prevention documented 49,400 suicide deaths in 2022, marking a 2.6% rise from the prior year. Men accounted for 75% of all suicides that year, with the steepest increases observed among non-Hispanic White and American Indian populations. Experts caution that while economic uncertainty and substance use have played roles, the reasons behind the trend require further study. Efforts to expand crisis intervention services and reduce stigma around seeking help continue to gain attention across policy and medical communities.

A new PostgreSQL extension called pg_vectorsearch provides an index type named PRISM that claims higher speed for approximate nearest neighbor queries than the IVFFlat and HNSW indexes used in the pgvector extension. The project remains in alpha and is not intended for production use; upgrading between prerelease versions may require dropping and recreating the extension and its indexes, according to the changelog.

PRISM, short for Partitioned Routing Index for Similarity Matching, is implemented in C and operates as a native PostgreSQL access method. It combines hierarchical centroid trees, RaBitQ quantization, and SIMD‑optimized distance calculations to achieve faster query response and index construction. In benchmark tests on a dataset of over 100 million vectors, PRISM matched or exceeded the recall of HNSW while using roughly one quarter of the memory, the developers reported.

The architecture follows a multi‑stage search flow. First, a centroid tree is traversed to locate the nearest clusters, leveraging the buffer cache to keep the tree resident in memory. Next, posting lists are scanned for candidate vectors, also benefiting from buffer cache and asynchronous I/O. A two‑stage RaBitQ filter then applies error bounds to eliminate vectors that cannot be nearest neighbors, reducing the number of items that must undergo full‑precision reranking. Finally, the remaining candidates are re‑evaluated against the original high‑precision vectors stored in the table, and the top k results are returned.

PRISM can be used in two distinct ways. Developers starting from scratch may store vectors in the extension’s own vec32 (float32) or vec16 (float16) column types, eliminating the need for pgvector altogether. Alternatively, existing tables that already contain pgvector’s vector or halfvec columns can install pg_vectorsearch alongside pgvector and create a PRISM index on those columns without altering the table schema or query logic. Both approaches support the same operator classes — <-> for L2 distance, <#> for inner product, and <=> for cosine similarity — and share the same tuning parameters.

Installation requires PostgreSQL 18 or later, a C11‑compatible compiler, and the Meson build system. Optional CBLAS libraries can accelerate RaBitQ encoding. After building the extension, users create the index with a statement such as CREATE INDEX ON items USING prism (embedding vec32_l2_ops); and adjust the nprobe setting to control the trade‑off between recall and latency. The default nprobe value of 40 offers roughly 95 percent recall; higher values improve recall at the cost of speed.

For tables that already contain a vector column, the extension can be added without disrupting existing indexes. After installing pg_vectorsearch, a PRISM index can be built on the same column, and queries using the <-> operator will continue to function. Performance comparisons can be verified with EXPLAIN plans, which typically show the planner preferring the PRISM index over alternatives such as HNSW when both are present.

Maintenance of a PRISM index involves periodic rebalancing of posting lists to prevent excessive growth. Inserts append to the appropriate posting list, and the prism_rebalance procedure can split oversized lists in place. Full reindexing is required only when schema changes or major version upgrades occur.

The project’s documentation includes separate sections on tuning, maintenance, architecture, and SIMD optimizations. It also cites academic work on binary quantization, incremental updates, and large‑scale ANN techniques that informed the design. The extension is released under the PostgreSQL License, and its development is tracked alongside related projects such as pgvector and pgvectorscale.