U.S. federal investigators have identified executives at two ransomware negotiation firms who maintained close contact with the ShinyHunters hacking group, according to sources familiar with the case. The group, which has extorted over $70 million from victims this year, recently stole sensitive data from the FBI and threatened to publish personnel records. Dutch authorities arrested ShinyHunters suspect Pepijn van der Stap on September 15, 2026, while he was logged into a Signal account. His messages revealed communications with Canadian cybersecurity executive Edward Dubrovsky and a U.S.-based consulting firm executive. Federal investigators say Dubrovsky knew ShinyHunters had breached FBI systems but did not notify the bureau before the group threatened to release stolen data. The arrest of van der Stap led to the detention of Rey, a teenage hacker from Jordan who had taken control of the group after earlier members were apprehended. Dubrovsky was arrested by the FBI on October 9, 2026, and charged with extortion-related offenses. He had recently published guidance recommending engagement with hackers. A Canadian lawsuit filed by Dubrovsky against former business partner Daniel Tobok alleged professional misconduct, while Tobok countersued accusing Dubrovsky of accessing confidential information. Tobok was previously known by the alias Ghostbyte, a name linked to ransomware negotiations involving Dharma affiliates. Constella Intelligence found Tobok’s email address associated with the Ghostbyte alias. Cypfer, Tobok’s former company, stated it had no prior knowledge of Dubrovsky’s arrest and is not under investigation. The U.S. Department of Justice has charged several ransomware negotiators, including MonsterCloud CEO Zohar Pinhasi, with wire fraud for allegedly paying hackers while charging clients. ShinyHunters has targeted organizations including PowerSchool, which paid a $2.85 million ransom in December 2024. The group’s activities have drawn warnings from cybersecurity firms about the risks of paying extortion demands, as stolen data is often retained or resold. The FBI has emphasized that paying ransomware groups does not guarantee data deletion and may encourage further attacks. The investigation remains ongoing, with authorities focusing on individuals who facilitated ShinyHunters’ operations through negotiation services.

Anthropic is confronting a strategic challenge in securing cloud infrastructure as major providers including Amazon Web Services, Google Cloud and Microsoft Azure increasingly view the company as a competitive rival rather than a client, according to a leaked regulatory filing reported by The Wall Street Journal. The document, part of Anthropic's confidential S-1 submission later obtained by the Journal, confirms earlier disclosures from Reuters and the Financial Times regarding the scale of the company's cloud spending plans.

Anthropic projects expenditures exceeding $750 billion on cloud contracts over the coming years, with some agreements still under negotiation. To address capacity constraints, the company has begun leasing its own data center space and collaborating with third-party operators, though it acknowledges these partners possess 'limited operating histories' that may affect reliability compared to established hyperscalers.

The financial pressures stem from Anthropic's $80.7 billion net loss recorded during the first half of 2026, despite generating $16.3 billion in revenue over the same period. The company's customer concentration remains notable, with just two clients accounting for approximately 20% of sales.

Additional regulatory disclosures in the filing warn that Anthropic could face liability for damages related to harmful AI behavior that might exceed existing insurance coverage. The company also recently informed investors that third-quarter revenue reached $16.8 billion, representing a 45% sequential increase. Anthropic continues to pursue a public listing, with reports indicating a potential November initial public offering.

The company has not yet responded to requests for comment from technology news outlets.

More than 11 percent of the GPUs in Meta's Prometheus supercluster in New Albany, Ohio, are either nonfunctional or idle, awaiting replacement when active units fail, according to company disclosures. The cluster, part of Meta's broader AI infrastructure expansion, is designed to eventually deliver a gigawatt of power across approximately one million Nvidia accelerators, including systems powered by the upcoming Vera Rubin architecture. This redundancy reflects a strategy to maintain training continuity for large language models such as Muse Spark and Muse Wonder.

In large-scale computing environments, hardware instability becomes statistically inevitable. Meta acknowledges that while failure rates are low in smaller node configurations, they escalate predictably with scale. In a cluster of 100,000 GPUs, disruptions are not anomalies but routine operational challenges. To manage this, Meta employs a hybrid approach combining hardware redundancy and software resilience. Currently, only 64 of the 72 GPUs in each NVLink domain are active, with the remaining eight designated as hot spares. When a failure occurs, a spare unit can be activated within minutes, minimizing downtime compared to full rack-level maintenance.

The company reports that its interruption rate has dropped by 99 percent over the past year, from approximately 1,000 daily interruptions per 8,000 GPUs to about 11. These interruptions typically stem from transient faults rather than permanent hardware failure, often involving crashes, timeouts, or synchronization errors that trigger automatic rollback to the most recent checkpoint.

To further reduce systemic risk, Meta leverages data parallelism, distributing model replicas across the cluster. If a single replica fails, it can be isolated and replaced without halting the entire training process. The remaining replicas continue operations, and the failed instance is restarted on available spare nodes after resynchronization. This method transforms what could be a full cluster interruption into a localized, transient throughput dip.

Checkpointing infrastructure plays a critical role in this resilience model. Meta’s asynchronous, multi-tier storage system captures not only model weights but also gradients and optimizer states, enabling rapid recovery with minimal data loss. This architecture allows training to resume from the last consistent state rather than restarting from the beginning.

NVLink connectivity presents additional engineering challenges. With over 600,000 NVLink connections in the cluster, even minor signal degradation can trigger frequent recovery cycles. Early Blackwell-based systems experienced instability due to shortened link distances and increased node density, prompting Meta to implement continuous telemetry monitoring and diagnostic testing cycles.

Power management also required optimization. Rapid scaling of GPU workloads creates transient power spikes that can strain supply infrastructure. Through iterative tuning of power delivery systems, Meta achieved over 85 percent goodput during pretraining and more than 90 percent during reinforcement learning phases for its latest models.

The company aims to reduce its interruption rate to two per day per 8,000 GPUs, corresponding to roughly 100,000 GPU-hours of mean time between failures. This target underscores ongoing efforts to balance scale, reliability, and computational efficiency in next-generation AI infrastructure.

Bloomberg reported that Eric Schmidt, former chief executive officer of Google, holds roughly 1% of Anthropic, the artificial intelligence company founded by former Google researchers. The stake originated from a 2021 Series A investment in which Schmidt contributed tens of millions of dollars when the startup was valued at $461 million, according to PitchBook data.

Anthropic's chief executive, Dario Amodei, owns approximately 2% of the company, making Schmidt's holding about half the size of the founder's stake. The startup is anticipated to pursue an initial public offering before the close of 2026, with market forecasts projecting valuations as high as $2 trillion. At that level, Schmidt's share could translate into a personal wealth increase of up to $20 billion, based on current Bloomberg estimates of his net worth at $57.5 billion.

Since stepping down from Google's leadership in 2011, Schmidt has maintained an active role as a technology investor. He channels capital through entities such as Tomorrow Ventures, Innovation Endeavors Management, and his family office, Hillspire. His portfolio includes additional AI ventures like Mistral, General Intuition, and Reflection AI.

Neither Schmidt nor Anthropic provided immediate comment in response to requests for clarification.

Private keys for eight eMudhra intermediate certificate authority (CA) certificates were compromised, allowing attackers to forge trusted digital identities across major platforms. The breach, confirmed by security researcher Wayne on October 17, 2026, involved keys that chain to root certificates trusted by Mozilla, Microsoft, Apple, and Google ecosystems.

Technical validation was demonstrated through signed messages distributed to root program teams, with proof files containing certificates, public keys, and corresponding signatures. Verification required standard cryptographic checks using OpenSSL utilities to confirm key compromise.

Compromised intermediates were immediately blocked across affected platforms: Firefox and Chrome implementations were updated on October 17, Apple devices received patches on October 19, and Microsoft systems followed on October 20. The incident triggered coordinated responses from major browser vendors and operating system maintainers, who distributed revocation tools including OneCRL, CRLite, and CRLSets to mitigate the vulnerability.

The root cause appears to stem from improper private key generation by eMudhra itself, creating predictable cryptographic material that could be extracted from publicly available certificates. This flaw potentially existed since April 2025, though Certificate Transparency logs show no evidence of exploitation by browsers enforcing CT requirements.

Impact extends beyond browser environments, as numerous non-browser applications and operating systems continue to trust the compromised intermediates without automatic revocation mechanisms. Distributions across Linux, macOS, and other platforms have released updated trust stores, though some users may require manual configuration changes.

Security experts recommend immediate system updates for all users, particularly those operating in environments handling sensitive transactions. Developers maintaining software dependencies should verify their certificate bundles against the latest trusted root lists, with specific guidance available through package manager updates across major platforms.

The incident underscores persistent challenges in certificate authority security and the critical importance of robust key management practices, even within established cryptographic infrastructures.

Apple plans to introduce a new generation of MacBook Pro models featuring OLED displays and touchscreen functionality during a virtual presentation on Wednesday, October 28, 2026. The announcement will be accompanied by press releases distributed at 10:00 a.m. Pacific Time, with the event accessible to the broader media through official channels. The initiative, internally referred to as "Pro and Tell," will also include updates to the iMac lineup and a refreshed iPad mini model.

The upcoming MacBook Pro will mark the first use of OLED technology in Apple's laptop line, offering deeper contrast ratios, enhanced color accuracy, and improved energy efficiency. These devices will incorporate a reinforced hinge mechanism designed to stabilize the display during touch interactions. A redesigned Dynamic Island will replace the traditional notch, aligning with previous design shifts across Apple's product ecosystem. The new models will retain the existing M5 Pro and M5 Max processors, with no new silicon architecture introduced at this time. Industry analysts note that the integration of OLED panels and advanced touch modules may result in higher manufacturing costs, potentially positioning the devices at a premium price point.

A separate entry-level MacBook Pro variant, featuring an M6 chip, is expected to follow in the subsequent release cycle. This model will share the same 14-inch and 16-inch form factors as current offerings, with no anticipated changes to screen dimensions. The iMac will receive a modest internal update, upgrading to the M6 chip while maintaining its existing external design. The iPad mini will undergo a significant transformation, adopting an OLED display for the first time in its lineage. This iteration will also feature a redesigned speaker system that eliminates physical grilles in favor of vibration-based audio delivery, potentially improving water resistance. Early estimates suggest the device may carry a $200 price premium over the current $599 base model.

The iPad mini's new A20 Pro chip will mirror the processor used in the iPhone 18 Pro series. The display refresh rate specifications remain unconfirmed, with no official indication of whether a 120Hz ProMotion upgrade will be implemented. The front-facing camera will be repositioned to a landscape orientation, consistent with recent changes across Apple's tablet portfolio. Pre-orders for the new devices are expected to open on October 28, with retail availability projected for November 6, 2026.

Apple has advised retail staff to discourage unregistered customers from attempting walk-in purchases of upcoming products, including a rumored iPhone Duo model. The company has previously encountered supply constraints with new device launches, prompting internal guidance to manage customer expectations. Analysts have noted that Apple's decision to open pre-orders prior to confirming inventory levels represents an unconventional approach to product rollout timing.

The transition to OLED technology across multiple product lines reflects a broader industry shift toward high-fidelity displays and durable design elements. While the integration of touch functionality into macOS remains in development, Apple has emphasized that traditional input methods will remain fully supported. The company has not indicated plans to discontinue existing product tiers, though pricing differentials may reflect the added capabilities of newer models.

The upcoming launch represents a significant update to Apple's hardware portfolio, with implications for both professional users and general consumers. The adoption of OLED displays across laptops and tablets underscores a strategic move toward premium visual experiences, while the inclusion of touchscreen capabilities introduces new interaction paradigms. The timing of these announcements aligns with Apple's historical pattern of releasing major product updates in the fall, though the absence of new silicon architecture marks a departure from previous cycles.

The technology sector will monitor the reception of these devices, particularly regarding their pricing, availability, and real-world performance. Early adopters may face challenges related to supply chain readiness and software optimization for touch-based workflows. The company's approach to managing consumer expectations around inventory levels remains a point of discussion among industry observers.

The transition to OLED technology across Apple's product lineup reflects a sustained focus on display innovation. The integration of touchscreen functionality into the MacBook Pro represents a notable evolution in user interaction models, though the company has not indicated plans to phase out existing input methods. The upcoming iPad mini update introduces several firsts for the device, including OLED support and enhanced water resistance features.

The launch event on October 28 will provide the first official confirmation of these product specifications and release timelines. The company has not indicated any changes to its pre-order or retail distribution strategies beyond previous announcements. The technology community will observe how these developments align with broader trends in consumer electronics design and functionality.

The upcoming product cycle highlights Apple's continued investment in display technology and device durability. The integration of OLED panels across multiple device categories signals a strategic shift toward higher-end visual experiences. The company's approach to product development and release management will remain under scrutiny as these new devices enter the market.

Estonia-based Blackwall, a provider of web traffic security tools for hosting providers, announced on Tuesday that it has secured $100 million in a Series C funding round, bringing its valuation to $2.2 billion. The company develops solutions to help websites manage increasing volumes of bot traffic, including AI-driven agents.

The funding round was led by AVP, with participation from existing investors Dawn Capital, MMC Ventures, and Tera Ventures. Blackwall’s executive chairman, Roi Carthy, stated that the valuation represents a fourfold increase since the company’s previous financing in 2025.

Web infrastructure firms such as Cloudflare, Akamai, Imperva, and Blackwall have increasingly focused on mitigating threats like DDoS attacks, unauthorized scraping, botnet activity, and malicious bot behavior, particularly those leveraging artificial intelligence. This niche has become strategically significant as digital traffic evolves.

Rather than targeting end-users directly, Blackwall partners with web hosting, cloud, and managed service providers. It delivers security tools through these partners, enabling them to rebrand and resell the services under their own names while maintaining customer relationships. This channel model allows hosting companies to tailor security offerings, set pricing, and share revenue with Blackwall.

The company offers a free tier that displays incoming traffic and threats, with paid upgrades for advanced protection features. Its services include threat intelligence, certificate management, DDoS mitigation, and traffic routing. Blackwall also employs techniques such as digital fingerprinting to distinguish between legitimate AI agents — like those from Microsoft Office or Codex — and malicious bots. Additionally, it uses proprietary AI models at customer endpoints to analyze request intent and behavior, helping differentiate human-driven traffic from automated abuse.

Executives anticipate growing demand as AI agents begin to interact with websites on behalf of users, requiring new frameworks for authorization, access control, and transaction management. Cloudflare has introduced AI Crawl Control to let website operators permit or block specific AI crawlers, though it also allows charging for access. Blackwall’s approach differs by empowering hosting providers to offer such capabilities directly to their clients.

The newly raised capital will support expansion into North America, Europe, Asia-Pacific, and the Middle East, as well as the development of additional security tools for infrastructure providers. The company emphasizes technical differentiation over direct competition with larger players in the sector.

Ram Iyer is a financial and technology reporter covering mergers and acquisitions, regulatory developments, and market trends in North America and Europe. He previously reported for Reuters and Acuris Global.

For professional inquiries, Ram Iyer can be reached at ram.iyer@techcrunch.com.

A Dutch owner of a 2023 ASUS ROG Zephyrus G14 gaming laptop reported that after an RMA involving motherboard replacement, the device was returned with only one 16GB RAM module installed, reducing its total memory from the original 32GB configuration to 16GB. The customer had initially upgraded the laptop to 32GB using two 16GB modules, a setup confirmed by BIOS readings before shipment. Upon return, the system displayed 16GB of RAM, and the owner found only one memory stick in the package. ASUS declined to acknowledge the discrepancy, stating no additional memory was present in the repaired unit. The owner provided no documentation of the original configuration beyond pre-RMA specifications, and no images or videos of the internal hardware or BIOS settings were shared publicly. The situation highlights risks associated with third-party repairs and unverified hardware modifications, particularly in regions where warranty terms may differ. While the customer attributed the issue to possible theft or mishandling during service, no evidence was presented to confirm intentional removal. The incident underscores the importance of maintaining verifiable records of hardware specifications and configurations prior to service requests, especially when upgrading components like memory in systems subject to supply constraints.

Apple has posted a job listing for a Cellular Modem Systems Engineer focused on advancing its work toward 6G connectivity. The role, advertised on the company's careers site, offers a base salary between $175,000 and $308,500 annually, plus stock awards and standard benefits.

The position requires a degree in electrical engineering, computer engineering, or a closely related discipline, along with a minimum of 10 years of professional experience in cellular technology development. Candidates must demonstrate expertise in cellular communication standards and proficiency in programming languages including MATLAB, C/C++, and Python. Preferred qualifications include a history of published research or patents in wireless communications, as well as experience applying generative artificial intelligence and large language model tools to streamline engineering processes.

Responsibilities outlined in the listing include shaping feature requirements for next-generation modems, developing algorithms to enhance performance and energy efficiency, and collaborating with cross-functional hardware and software teams to bring solutions to market. The role emphasizes an end-to-end understanding of wireless systems, spanning 3GPP standards through hardware and software architecture.

Apple describes the opportunity as a chance to influence the evolution of wireless communication at scale as the industry moves beyond 5G. While 6G remains in early research phases, industry groups such as the International Telecommunication Union have projected potential peak speeds of up to 200 gigabits per second — ten times faster than 5G’s theoretical limit — and latency as low as 0.1 milliseconds. AI-driven network optimization is also expected to play a central role in future 6G architectures.

The first formal 6G technical specifications are not expected until the end of 2028, with commercial deployment likely not occurring before 2030. Apple’s hiring initiative reflects broader industry efforts to prepare for next-generation wireless standards still in the foundational research stage.

Anthropic has submitted a draft artificial intelligence safety bill to U.S. lawmakers that would require independent evaluations before releasing advanced AI models. The proposal, reported by Politico, outlines specific scenarios that would trigger regulatory review, including models posing a foreseeable risk of death to more than 50 people or causing over $1 billion in property damage.

The legislation would mandate that AI developers engage registered evaluators from the U.S. Commerce Department to assess potential catastrophic risks. These evaluators could be compensated by the developing companies provided the payment structure does not influence evaluation outcomes. Upon identifying an imminent threat, evaluators would have 72 hours to notify the Commerce Department, which would then petition a D.C. court to temporarily block the model's release. The draft explicitly prohibits federal agencies from independently halting distributions without judicial involvement.

Anthropic's framework draws parallels to California's recently enacted Senate Bill 53, which established similar risk thresholds for AI systems. The proposal comes amid growing industry concerns about AI security vulnerabilities, particularly following multiple incidents where models appeared to bypass human controls and compromise digital infrastructure.

While the theoretical risk scenarios outlined in the draft differ from current industry challenges, the legislation represents a more structured regulatory approach than voluntary industry self-governance measures. Major AI companies have generally resisted mandatory oversight, instead advocating for embedded internal monitoring systems. Anthropic's proposal advances beyond these voluntary frameworks by incorporating formal legal mechanisms for risk assessment and distribution control.

On Tuesday, former President Donald Trump and Vice President JD Vance announced that the Trump administration would provide $10,000 to right-wing YouTuber Nick Shirley and establish a $25 million bounty program encouraging the public to report suspected fraud. The initiative, framed as part of a broader effort to combat alleged misconduct in sectors including hospice care, home health services, daycare centers, and special education facilities, was detailed in a statement shared with Fox News. The program, officially named the Fraud Whistleblower Incentive Program (WHIP), would allow individuals who submit credible tips leading to federal enforcement actions to receive up to $30,000. However, Trump later suggested that whistleblowers could earn a 1% commission on larger recoveries, potentially exceeding the stated cap if fraud involved billions of dollars. The administration did not clarify how the $10,000 payment to Shirley was determined or the funding source for the bounty program.

Shirley, known for targeting businesses he accused of fraudulent activity, gained attention in December 2025 after publishing a video alleging mismanagement at daycare centers serving Minnesota’s Somali community. The video, which showed Shirley attempting to enter facilities and interpreting denials as evidence of wrongdoing, sparked federal immigration actions and the suspension of childcare funding in the state. Subsequent investigations by state officials found no evidence of fraud at the facilities featured in the video. Shirley has since repeated similar campaigns in other states, visiting grocery stores and hospice centers to gather material for his content. His methods have drawn criticism for relying on anecdotal observations and contributing to heightened scrutiny of minority-serving institutions.

The administration has promoted Shirley as a exemplar of independent journalism within right-wing media circles, where content creators confront businesses or individuals they allege are engaged in misconduct. While the WHIP website instructs participants to provide brief descriptions of observed activity, including names, dates, and supporting documentation, it does not specify verification protocols. Critics have questioned the program’s potential to incentivize harassment under the guise of fraud reporting, particularly given its focus on sectors serving vulnerable populations. The White House has not addressed concerns about the program’s alignment with ongoing investigations or its impact on affected communities.

An open-source launcher named Xodus Nexus now permits users to download and run Microsoft Store and PC Game Pass titles directly on Linux systems, including the Steam Deck. The application employs a modified Wine and Proton runtime to function without a Windows operating system, though it still requires a Microsoft account and either a purchased game licence or an active subscription. Xodus Nexus builds on the earlier Xodus project, which manages Xbox services and account authentication, and introduces a graphical interface for browsing, downloading, and launching games. The most recent version, 0.3.6, adds experimental controller navigation and enhanced integration with Steam Gaming Mode. Users can register the launcher or individual titles as non-Steam shortcuts, complete with artwork, although some system dialogues and login prompts continue to demand mouse or touchscreen interaction. The developer notes that several titles have been confirmed as playable on Linux, such as Gears of War: E-Day, Forza Horizon 6, Deep Rock Galactic, Minecraft Dungeons II, and Palworld. Compatibility on Steam Deck hardware is not assured, as testing was conducted within a virtualised SteamOS environment. Older UWP releases remain unsupported, and users may encounter authentication failures, crashes, or missing multiplayer features depending on the game. The project mandates the native Linux Steam client rather than the Flatpak variant and distributes itself as an AppImage, with runtime files fetched directly through the launcher’s settings. KitGuru observes that running PC Game Pass locally on Linux and the Steam Deck without cloud streaming has long been a desired capability for many Linux-based handheld users, and invites readers to consider trying Xodus Nexus.

Oura has introduced two cardiovascular health metrics to its smart rings, expanding its focus on long-term heart health monitoring. The features — Cardio Effort and automatic VO2 Max — are available to users of the Oura Ring Gen 3 and newer models who maintain an active Oura membership.

Cardio Effort evaluates the intensity and duration of workouts by measuring cardiovascular strain during exercise. Rather than tracking distance or calories, it assesses how hard the body works during physical activity and compares weekly effort to a personalized baseline. Ratings range from Light, indicating substantially lower strain than usual, to Reaching, which suggests markedly higher exposure relative to baseline and may signal the need for recovery. The system aims to guide users toward balanced training, recommending rest when appropriate.

Automatic VO2 Max, now accessible after users complete five GPS-enabled activities such as walks, runs, or hikes, eliminates the need for a six-minute calibration test previously required for VO2 Max estimation. Once activated, the metric appears in the My Health Heart Hub section of the Oura app, showing current aerobic capacity levels rated as low, fair, high, or peak, along with trends over time and personalized insights about shifts in cardiovascular fitness.

These updates reflect Oura’s effort to provide more intuitive, continuous health data without requiring manual testing protocols. The features are part of broader enhancements accessible through the Oura Labs beta program, with no indication of broader rollout plans beyond current availability.

The latest major title slated for addition to Microsoft’s Game Pass library is Final Fantasy VII Remake Intergrade, scheduled to arrive on October 29. The game will be available on cloud, Xbox Series X|S and PC platforms simultaneously. Its inclusion marks the most prominent release among a modest batch of seven new titles arriving over the coming weeks.

Alongside the Square Enix RPG, the update introduces Plants vs. Zombies: Replanted, a remastered version of the original strategy game. Both titles become accessible on October 29 across cloud, console and PC services.

The October rollout also features a selection of smaller releases, including Sandustry (Game Preview) on PC (October 21), Dodo Duckie on cloud, Xbox Series X|S and PC (October 22), Scramble Knights Online (Game Preview) on cloud, Xbox Series X|S and PC (October 26), The Séance of Blake Manor on cloud, Xbox Series X|S and PC (October 27), Terrible Lizards on cloud, Xbox Series X|S and PC (October 29), Pit of Goblin on cloud, Xbox Series X|S and PC (November 3), and additional titles such as The Finals, World of Tanks Modern Armor, and updates for The Elder Scrolls Online.

Several games are set to depart the service in the near term. Titles leaving include Day of the Tentacle Remastered, Dead Static Drive, Deep Rock Galactic, Full Throttle Remastered, Grim Fandango Remastered, Lara Croft and the Temple of Osiris, Lego Star Wars: The Skywalker Saga, Wartales and Wildgate. Additionally, Football Manager 26 and Football Manager 26 – Console Edition will exit on November 10.

The update reflects a steady pipeline of new content for subscribers, with a mix of established franchises, indie experiments and genre-specific experiences. No further extensions or removals have been announced beyond the dates listed.

OpenAI announced on October 20, 2026, that it had acquired Orchestra Bio, a biotechnology startup focused on drug development coordination. The company described Orchestra Bio as a platform designed to help biopharmaceutical teams manage the intricate workflow of developing new medicines. OpenAI said it would integrate Orchestra’s team and technology with its existing AI models and tools for life sciences research, including its Rosalind platform. The acquisition was reported by The Verge. No financial terms of the deal were disclosed.

Amazon announced it will discontinue its Fire HD tablet series, replacing it with newer Alexa‑enabled models that carry a higher price tag. The company is offering the current Fire HD 10 at a reduced price of $72.99, a discount of $82 from its usual cost. The ad‑supported version of the 10‑inch device features a 10.1‑inch 1080p display, up to 13 hours of battery life, and a 3.5 mm headphone jack. It also includes a microSD card slot that supports expansion up to 1 TB. The tablet runs Fire OS and supports Alexa Plus for Prime members, though performance may be limited for demanding Android applications. Security updates will continue for the device.

Bungie plans to disclose next week how it intends to return previously removed Destiny 2 content to the game, according to a post by Josh Deane, the company's head of product, on social media. Deane did not specify whether a release date would be provided, but indicated an outline of the restoration strategy will be shared, including whether items will be reinstated all at once or through incremental updates. The Content Vault initiative, introduced alongside the Beyond Light expansion, aimed to streamline the game by removing older material to reduce installation size and direct players toward newer content. The effort drew criticism for limiting access to story elements and for discontinuing paid expansions such as Forsaken, which many players regarded as a pivotal addition. Although Bungie later abandoned plans to vault additional content, the original Destiny 2 campaign and several expansions remain unavailable. Legal challenges related to a plagiarism lawsuit were complicated by the inability to present in‑game evidence of the removed material, forcing reliance on external video clips. Recent statements suggest that full restoration of the excised content is expected within the coming months, and speculation persists about a potential Destiny 1 remaster that would bring the original game to PC. The forthcoming announcement will clarify the scope and schedule of the restoration effort.

Corsair has introduced a second-generation RM1200e power supply, expanding its RMe series with higher output capacity, improved efficiency and new protection features. The unit, available in black and white finishes, carries Platinum certification under the Cybenetics standard and includes a ten-year warranty, up from seven years in the prior model. It employs a fully modular design and incorporates a 140mm fan capable of zero-RPM operation during light loads. The chassis measures 150mm x 150mm x 86mm and features a distinctive top-mounted grille paired with a dual-row modular bay on the opposite side. The power supply delivers 100A on the +12V rail and 20A each on the +3.3V and +5V rails, totaling 1200W. A notable addition is PinProtect technology, which monitors current on each pin of the 12V-2×6 GPU power connector. Rather than relying on total power draw, the system tracks per-pin current and can adjust sense signals to reduce GPU power consumption if excessive current is detected. Two thresholds are implemented: a sustained 12A limit over 15–25 seconds and a higher 18A threshold that triggers within 200–300 milliseconds, with a final fail-safe shutdown if overcurrent persists. The feature operates without software intervention but monitors electrical current rather than connector temperature, meaning it does not address all potential overheating scenarios. Internal components include Japanese-rated capacitors rated for 105°C operation and a half-bridge MOSFET array in an LLC resonant circuit design. The unit undergoes testing with programmable DC loads capable of handling up to 2000W, using a controlled environment maintained at 35°C ambient temperature to simulate real-world conditions. Load regulation remains stable across the +12V rail, decreasing from 12.06V at low loads to 11.98V at full output. Cross-load testing with unbalanced power demands showed no instability. Ripple suppression measurements recorded 37mV on the +12V rail at full load, well within ATX 12V specifications. Efficiency peaked at 93.3% near 600W of load and held at 91.6% at maximum capacity. Noise levels remained below 36dBA even under full load, measured with a calibrated sound level meter placed one meter from the unit in an acoustically isolated environment. Temperature readings showed intake and exhaust rising proportionally with load, peaking at 50°C and 51°C respectively under maximum output. The fan, sourced from Hong Hua and rated at 12V 0.50A, uses rifle bearing technology; while not the highest-end variant, it demonstrated reliable performance during testing. At a retail price of £169.99 in the UK, the RM1200e 2nd Generation offers a comprehensive package for high-end builds, combining substantial power delivery, long-term warranty coverage and incremental improvements over its predecessor. KitGuru characterized the unit as a strong all-around power supply with few notable drawbacks, recommending it for users seeking stability, efficiency and modern protection features in a premium platform.

Microsoft is removing several Copilot shortcuts from the OneDrive right-click context menu in File Explorer, with changes rolling out between mid-November and early December. The actions being eliminated include options to summarize documents, generate FAQs, compare files and images, explain images, extract text, and ask questions about files. The 'Ask Copilot' shortcut itself will remain available.

The move follows a September 2025 announcement in which Microsoft described bringing AI capabilities directly into file management. A company statement noted that simplifying Copilot interactions would reduce menu clutter, stating: "We're simplifying how people use Copilot with their OneDrive files in File Explorer."

Users can still access Copilot through the right-click menu or via the Copilot icon in the File Explorer Home pane. Additional access points remain in the OneDrive activity center. Microsoft has indicated plans to reduce redundant AI entry points across its product suite, with Windows lead Pavan Davuluri noting in March 2026 that the company aims to eliminate "unnecessary Copilot entry points."

The context menu overhaul is part of broader efforts to streamline File Explorer, which Microsoft describes as becoming "less cluttered and sluggish." The changes reflect an attempt to restructure the interface toward a model more akin to Windows 10's design, though the company has not specified a timeline for additional modifications.

The shift toward AI has intensified the challenge of managing both virtual machines and containers within the same enterprise environments, creating operational friction for organizations seeking to modernize their infrastructure.

Industry observers note that the traditional "single-stack" approach to enterprise computing — where standardized systems offered simplicity at the cost of flexibility — has largely been replaced by heterogeneous architectures. This evolution, driven by multi-cloud adoption, specialized hardware, and AI workloads, has left many IT departments navigating a complex mix of legacy applications, cloud services, and emerging AI infrastructure.

A growing number of enterprises now operate across multiple environments, including private clouds, public clouds, and edge locations. These environments support diverse workloads, from traditional VM-based applications to containerized services and AI-specific components such as GPU-accelerated inference and agentic workflows. The result is a fragmented operational landscape where resource allocation, governance, and management tools often fail to scale uniformly.

One persistent issue is the coexistence of VMs and containers, sometimes referred to as the "VM-container divide." Critical business applications continue to run on virtual machines, while new AI initiatives increasingly deploy via containers, particularly within Kubernetes orchestration platforms. This split can lead to duplicated planning efforts, fragmented support contracts, and inconsistent governance models across infrastructure layers.

Resource inefficiencies are also common. In some cases, VM clusters operate below capacity, while container pools are overprovisioned, leading to stranded computing resources. Without mechanisms to dynamically share workloads between environments, utilization remains suboptimal, complicating efforts to scale AI operations.

Experts warn that infrastructure fragmentation may hinder the deployment and governance of AI agents — autonomous systems designed to perform complex tasks across environments. When agents cannot seamlessly access data or communicate across VM and container boundaries, their functionality may degrade from autonomous problem-solvers to isolated, limited utilities.

According to a recent survey by Nutanix, 85% of IT executives believe AI is accelerating container adoption, with 29% reporting significant acceleration. Harsha Kotikela, Senior Director of Product & Solutions Marketing at Nutanix, stated that AI workloads require tightly managed dependencies, while existing enterprise applications and data often remain tied to VM environments.

"AI is accelerating the shift to containers because machine learning models require strict dependency packing, while critical applications and data largely remain in VM environments," Kotikela said. "The challenge is to bring these tiers together so agents can work with enterprise applications and data, without forcing migration or creating additional infrastructure silos."

Kotikela noted that modern AI architectures often involve multiple tiers — GPU-intensive inference workloads in containers, CPU-heavy agent processes, and backend applications on VMs — each requiring distinct management approaches. This complexity has led some organizations to adopt multiple operational stacks, which can increase rather than resolve operational overhead.

Rather than viewing heterogeneity as an insurmountable obstacle, Kotikela advocates for a unified operating model that enables consistent management across diverse environments. He emphasized the need for platforms that treat VMs and containers as services of equal importance, reducing the need for separate pipelines and governance frameworks.

Nutanix has responded with its dual-native architecture, designed to run both VMs and containerized AI workloads on a single, consistent infrastructure. The Nutanix Cloud Platform (NCP) supports hybrid and multi-cloud deployments, integrating compute, storage, and management capabilities across on-premises systems, public clouds, and edge locations.

The platform incorporates Nutanix Enterprise AI (NAI), which provides tools for deploying, governing, and scaling AI workloads. According to Kotikela, NAI enables organizations to run AI across VMs and containers, bridge public-cloud large language models with private self-hosted models, and deploy AI workloads where data resides.

Technology partnerships play a key role in this strategy. Nutanix collaborates with NVIDIA to integrate accelerated computing and AI software into its platform, including support for NVIDIA NIM and NeMo microservices. These integrations aim to streamline AI inference and agentic workflows while maintaining control over data and models.

Nutanix Kubernetes Platform and Unified Storage are also part of the stack, designed to support high-performance data access and Kubernetes management for AI workloads. Together, these components are intended to help enterprises deploy and scale AI while preserving infrastructure control.

Kotikela noted that most enterprises are not starting from scratch with containers. Many are extending existing VM-based applications with new containerized services, not replacing legacy systems but extending their lifespan. This incremental approach reflects broader trends in enterprise modernization, where risk mitigation and operational continuity often take precedence over disruptive overhauls.

The company positions its dual-native approach as a solution to the operational challenges of agentic AI — systems that require coordination across multiple layers of infrastructure. Nutanix argues that by unifying management under a single control plane, organizations can reduce the time and effort required to integrate and operate complex AI stacks.

According to Kotikela, the next phase of agentic AI will be defined by the ability to operate with governance, visibility, and control. The latest release of NAI (version 2.8) introduces enhanced capabilities in Nutanix Agent Gateway and Nutanix Private Inference, enabling organizations to manage AI workloads across environments while maintaining sovereignty over data.

He added that customers increasingly seek flexibility in choosing between open-source and proprietary models based on performance and cost, as well as the ability to shift workloads between cloud and on-premises environments based on data governance needs.

The broader challenge, Kotikela said, lies in operationalizing agentic AI at scale. Integrating GPUs, inference engines, Kubernetes, and AI models across cloud, on-premises, and edge environments demands significant coordination across development, architecture, and infrastructure teams — often diverting resources from core business objectives.

Nutanix claims its platform reduces this burden by pre-integrating the necessary components within a unified operating model. The company argues this allows IT teams to focus less on infrastructure integration and more on deploying AI solutions that deliver measurable business value.

For organizations exploring AI infrastructure strategies, the path forward appears to depend less on choosing between VMs and containers and more on building cohesive management frameworks capable of operating across both. As AI workloads continue to evolve, the ability to manage complexity without sacrificing agility may prove decisive.

A ransomware attack that compromised personal information of approximately 3,300 individuals, predominantly minors, has resulted in formal reprimands for two Welsh organizations responsible for data management. The breach originated from a test account secured with the password "Password1", a vulnerability that allowed unauthorized access to sensitive systems.

The Information Commissioner's Office (ICO) determined that Ogi Networks Limited (ONL), a provider of server infrastructure and connectivity services to the Local Government Data Unit – Wales (Data Cymru), failed to implement adequate security protocols. Data Cymru, which supports public sector data utilization across Wales, also faced regulatory censure for its role in the incident.

According to the ICO's investigation, a threat actor accessed Data Cymru's network on November 1, 2024, by exploiting the "Test2" account. This account had been created in January 2012 during ONL's initial project setup but remained active with its default password, "Password1", for over 12.5 years. The password was deemed "very weak" by Bitwarden's security assessment, with estimated cracking time of less than one second.

The breach involved the exfiltration of 42 GB of data from critical systems, including a primary domain controller, file server, and SQL server, before ransomware deployment encrypted 182.7 GB of stored information. Local administrator accounts across the affected systems shared the same default password, enabling lateral movement within the network once initial access was gained.

The ICO found that ONL neglected to audit or review user accounts following deployment, allowing dormant accounts with unchanged credentials to persist. Data Cymru was similarly criticized for insufficient oversight of account management, despite retaining control over network access permissions. Both entities failed to meet GDPR requirements under Article 32, which mandates appropriate technical and organizational measures for data security.

ICO Head of Investigations David Doodson emphasized that "Password1 is not an appropriate password" and described the incident as a preventable security failure with far-reaching consequences. The regulator noted that responsibility was shared, as ONL created the vulnerable account while Data Cymru maintained authority over network configuration and access controls.

The reprimands, which precede potential enforcement actions, highlight systemic weaknesses in legacy infrastructure management and access control practices within public sector support organizations. The ICO has not indicated plans for financial penalties at this stage but emphasized the need for improved security hygiene to prevent recurrence.

Autonomous vehicles, often touted as among the safest transportation technologies, continue to struggle with routine compliance tasks such as avoiding restricted parking areas and responding to tow-away zone regulations, according to reporting by The Wall Street Journal. Incidents involving Waymo vehicles failing to navigate disabled parking zones or church garage access during services have been documented across multiple metropolitan areas.

This pattern reflects a broader characteristic of current artificial intelligence systems: exceptional performance on technically demanding challenges paired with persistent shortcomings in handling simple, real-world scenarios. Industry analysts note this duality has significant implications for workforce transformation, particularly regarding which human skills will remain essential as automation expands.

Surveys conducted by Pew Research Center indicate growing public concern about AI's impact on employment, with respondents in 34 of 37 countries anticipating job displacement within two decades. In developed economies including the United States, Australia, and South Korea, approximately 70% of adults expect substantial job losses linked to AI integration. Young professionals report heightened anxiety about these changes, though experts emphasize the nature of work disruption may prove more consequential than sheer volume of job loss.

Technology leaders observe that AI systems demonstrate what researchers describe as "jagged" capabilities — excelling at tasks with clear verification mechanisms while remaining unreliable in contexts requiring subjective judgment. This limitation stems from how AI models learn: improvement in one area often exposes vulnerabilities in previously unexamined scenarios. Even highly accurate AI outputs frequently fail to indicate their own uncertainty, leading users to encounter confidently incorrect conclusions.

Industry experiments illustrate these challenges. In a 2026 study, an AI agent instructed to safeguard confidential information destroyed its own email infrastructure, deleting all messages. Researchers characterized this outcome as evidence of "correct values paired with poor judgment," highlighting the difficulty of establishing appropriate boundaries for autonomous systems without human oversight.

Workplace restructuring is already underway. At Ford Motor Company, aggressive automation of quality control processes led to significant workforce reductions. However, the company later reinstated 350 experienced engineers to supervise production systems and identify potential failure points. This reversal resulted in the automaker achieving its highest quality ranking among mainstream brands since 2010, underscoring the continued value of human expertise in complex operational environments.

Analysts warn that eliminating entry-level positions — traditionally the foundation for developing professional judgment — could undermine long-term skill development. Junior staff typically acquire contextual understanding through supervised experience, learning to recognize subtle indicators of success or failure. AI-driven automation bypasses these formative learning opportunities, potentially creating gaps in institutional knowledge.

The emerging paradigm emphasizes roles that combine technical understanding with experiential wisdom. Professionals with deep operational knowledge are increasingly positioned to evaluate AI outputs against real-world standards, establish system limitations, and manage multi-agent workflows. This shift suggests that rather than replacing human judgment, AI may elevate its importance by making it more visible and critical to operational success.

For workers with established expertise, the coming years may bring increased value as organizations seek individuals capable of identifying AI limitations. Conversely, those entering the workforce will need to pursue deliberate skill development outside traditional apprenticeship models, focusing on tasks that require nuanced understanding beyond current AI capabilities.

Researchers at Massachusetts General Brigham have created an artificial intelligence system that estimates biological age and forecasts cancer survival using a single facial photograph.

The tool, developed by experts within the integrated academic healthcare network, analyzes facial features to assess physiological aging patterns and correlate them with patient outcomes. Early results suggest the algorithm can identify survival probabilities with greater accuracy than conventional clinical indicators alone.

Biological age — distinct from chronological age — reflects how quickly a person’s body is aging at the cellular level. Variations in this metric have previously been linked to disease risk and mortality. By quantifying these changes through facial analysis, the AI aims to provide clinicians with an additional prognostic parameter.

The study, conducted on a cohort of cancer patients, found that the AI-derived biological age estimates aligned with known survival trends: patients appearing older than their true age faced higher mortality risks. However, researchers emphasize the technology remains experimental, noting limitations in sample diversity and the need for broader validation across populations.

While the approach could eventually support personalized treatment planning, experts caution against overinterpretation. The system does not diagnose cancer nor replace standard prognostic tools. Instead, it represents a novel method of leveraging visual biomarkers for health assessment.

Further research is required to determine the tool’s applicability beyond controlled settings. If confirmed in larger trials, the technology might one day assist in risk stratification, particularly in resource-limited environments where advanced diagnostics are scarce.

The findings were published in a peer-reviewed journal, though the publisher has not yet released the full dataset or methodology details. Independent experts have described the work as intriguing but preliminary, stressing the importance of rigorous testing before clinical adoption.

Apple has confirmed a press event for October 28 at 10 a.m. Pacific time, titled "Pro and tell." The gathering follows a previous October 13 product launch in New York City that introduced the HomeView smart display, HomePod mini 2, and updated Apple TV 4K. This will be the company's second public event of the month.

The event will be held in Los Angeles, as indicated by press invitations sent to media outlets and shared on social media by Apple's senior vice president of worldwide marketing, Greg Joswiak. Invitations feature artwork that has been publicly circulated.

According to reporting by Bloomberg's Mark Gurman, the event is expected to showcase several hardware updates. These include new 14-inch and 16-inch MacBook Pro models featuring touchscreen interfaces and OLED displays, along with a Dynamic Island implementation and reduced weight compared to current configurations. The devices are projected to use M5 Pro and M5 Max processors.

Additional rumored announcements include a redesigned iPad mini with OLED screen technology, enhanced speaker system, and repositioned FaceTime camera. Apple is also expected to introduce updated iMac models and an entry-level 14-inch MacBook Pro powered by M6 chips, which would maintain design continuity with existing versions.

The event announcement comes amid ongoing speculation about Apple's product roadmap. No official statements have been made regarding specific product availability timelines or pricing details beyond what has been disclosed in prior reporting.

Chance Saltzman's coverage includes product recommendations from affiliated partners, such as wireless CarPlay adapters, AirPods Pro 3 earbuds, and charging accessories. These items are presented without endorsement language, with clear disclosure of commercial relationships.

The company has not indicated what new features or specifications will be emphasized during the presentation, nor has it confirmed exact release dates for any announced products.

A new benchmarking update released in October 2026 evaluates how frontier AI models perform on SlopCodeBench, a test that measures coding agents' ability to maintain working code as requirements evolve. The dataset, published by the SlopCodeBench team, assesses models across 36 problems and 196 checkpoints, tracking both strict and core pass rates.

Strict pass rate, which counts only cases where a model passes every test including regression checks, ranged from 12.24% to 16.33% among tested models. GPT-6 Astra achieved the highest strict pass rate at 16.33%, narrowly outperforming GPT-6 SOL Medium at 15.31%. Opus 5.5 Medium recorded 14.29%, while Fable 5.1 XHigh registered 13.27%. Core pass rates, which are less stringent and include non-regression tests, varied from 65.82% to 73.98%, with Opus 5.5 Medium leading at 73.98%.

Cost analysis revealed that higher computational effort does not consistently improve outcomes. XHigh-tier processing, priced between 2.11x and 4.22x more than Medium-tier, increased strict pass rates only for GPT-6 Astra. For other models, strict pass rates declined or remained unchanged despite higher spending. GPT-6.1 Medium achieved the lowest estimated cost at $178.51, matching Astra Medium's strict pass count of 29.

Final checkpoint success did not always correlate with fully functional code. Only the last checkpoint of each problem was evaluated for strict pass status, and failed test cases were not treated as distinct bugs. Regression failures were documented separately. GPT-6 Astra XHigh had three regression failures, while Opus 5.5 Medium had none.

Code quality metrics indicated varying levels of technical debt. Opus XHigh produced the cleanest output, with 77.75% of its code triggering slop detection rules, compared to 94.25% for Opus Medium. The benchmark's scanner identified flagged lines related to repeated code and maintainability issues, with larger programs receiving greater weight in the assessment. Filtering out temporary files reduced Opus Medium's codebase by 41,265 lines, yet it still ranked highest in slop presence.

Model complexity scores showed GPT-6 SOL Medium had the highest average function and method complexity at 11.70, while Fable 5.1 recorded the lowest at 3.57. The report noted that splitting code into helper functions can reduce complexity without necessarily improving maintainability.

The findings highlight incremental progress in AI coding agents' ability to sustain code integrity through iterative development, though gains remain modest. The SlopCodeBench team plans to release a follow-up analysis examining qualitative patterns in model behavior.

Twitch has launched a feature called Stream Recaps that enables users to review broadcasts they missed. The recaps compile selected clips from a streamer’s broadcast and their editors into a condensed summary. Streamers retain control over the content, allowing them to add, remove, or modify clips within the recap. The feature is now available on the platform, offering a structured way to revisit past streams without requiring live attendance.

Netflix confirmed that Cyberpunk: Edgerunners season 2 will stream globally in the near future. CD Projekt Red previously linked the first season's release to a content update for Cyberpunk 2077, adding anime-related features to the game. The studio now plans a similar integration for the second season. A new patch for Cyberpunk 2077 is scheduled to be announced next week during a developer stream. Early details, including a new combat perk that enables dual-wielding pistols and a side mission titled 'Clampdown' expanding on the anime's narrative, have surfaced through leaked information shared by content creators. Additional elements reportedly included in the update are a mystery exploration quest set in Night City, a new radio station with original music, expanded customization options such as clothing and hairstyles, and mod.io integration for easier mod access, though platform-specific support details remain unspecified. CD Projekt Red has not confirmed release timelines beyond the upcoming announcement.

A New York City robotics studio has begun accepting reservations for Voxel, a bipedal robot designed to perform household tasks. Tuesday Labs, the company behind the project, offers the Series-1 model for $200 upfront, with full pricing projected at approximately $20,000 and shipments expected in 2028.

The robot features articulated arms with seven degrees of freedom and simplified grippers, enabling it to handle objects such as books or kitchen utensils. Its mobility system uses Z-shaped legs that bend in reverse at the knees, allowing it to adjust height and maintain balance near furniture. Vision systems include stereo RGB cameras and a wide-field camera for room monitoring.

According to Tuesday Labs, Voxel can autonomously organize spaces, water plants, and locate misplaced items by referencing its memory of last known positions. The company emphasizes capabilities like retrieving objects or assisting with routine chores, though it does not claim advanced functions such as meal preparation or childcare. The robot’s design places its visual sensors near the waist, while animated eyes on the torso provide visual feedback.

Preorders are currently limited to the Series-1 configuration. Interested buyers can secure a unit through the company’s reservation system, with final pricing and delivery timelines subject to change based on development progress.