Federal officials are asking whether the acquisition would hand Nvidia a chokepoint over how AI models reach developers, adding to a widening review of the chipmaker's dealmaking.
An autonomous system called Aardvark and a related open-source repair effort have surfaced nearly 2,000 software flaws this year, but only a quarter have been fixed — exposing a widening gap between finding bugs and closing them.
Blockchain investigators say the theft used social-engineering tactics researchers warned about for years, as Binance scrambles to trace the money and rebuild trust.
A revised IPO filing says future Claude models could soon cross a threshold the company defines as dramatically accelerated, AI-driven progress — and admits it has missed such leaps before.
The suits accuse Brave of building a 30-billion-page index that publishers can't block, then supplying scraped content to AI companies including Anthropic, Perplexity and xAI.
A fresh wave of Apple threat notifications reached members of Germany's parliament and prominent figures in roughly 99 countries, reviving scrutiny of the mercenary-spyware industry even in nations with mature cybersecurity institutions.
Attackers remotely pushed refrigeration units into defrost mode overnight, spoiling food at chains including Delhaize, Carrefour and IKEA — the second such wave to hit European grocers in a month.
The Suncatcher prototype, a space-based AI computer launched this month, passed its first commissioning tests by running a Gemma model that greeted engineers with a poem about orbit.
A federal judge rejected xAI's bid to dismiss a Clean Air Act suit over dozens of unpermitted gas turbines powering its Colossus supercomputer, in a ruling that lands amid a national backlash against data-center buildouts.
After more than two years without Apple Intelligence in its biggest overseas smartphone market, Apple will let Chinese iPhone users route Siri, Writing Tools and Visual Intelligence requests through Alibaba's homegrown AI model.
A custom exploit chain combining a PHP flaw with a Linux kernel bug let attackers seize root access on Palo Alto Networks firewalls across the US, Japan, Taiwan and South Korea.
Attackers are remotely rebooting unpatched ASA and FTD firewalls by exploiting a flaw in Cisco's SSL VPN service, the company's second such zero-day in three months.
A 36-state settlement requires General Motors to get explicit consent before collecting driving data, capping a two-year fallout from its discontinued Smart Driver program.
A supply-chain compromise of Anthropic's official VS Code extension on Cursor's marketplace tried to plant SSH backdoors on developer machines before the companies pulled it — and neither has said how attackers got in.
Social media posts show display lines and crease cracks on the iPhone Duo just days after launch, as analysts say Apple's handling of the issue will shape confidence in foldables beyond its own sales.
Binance says a compromised engineer's laptop let an attacker submit unauthorized withdrawals, in a hack analytics firms say bears the fingerprints of North Korea's Lazarus Group.
The Russian president said he is in "close" contact with Musk over satellite orbits, even as Starlink and Russia's fledgling Rassvet network remain wildly mismatched in scale.
The regulator has recommended photo ID checks, bank-based verification and face-age estimation as the UK prepares to lock out under-16s from Instagram, TikTok and eight other platforms.
The proposed "AI company" structure would let an autonomous agent sign contracts, hold property and be sued in its own name inside a supervised state sandbox, but critics say reports of agents acting unpredictably undercut the case for giving them legal cover.
Third-party Android app stores outside the US can now apply for the same one-click install and silent-update privileges Google's own Play Store enjoys, a voluntary alternative to the court-ordered access US stores already have after Google's loss to Epic Games.
The new world model generates interactive 3D environments frame by frame as users explore them, becoming the first such system to produce native audio, and will roll out first to Google AI Ultra subscribers and select developers.
The company paused deployments and urged all customers to rotate their secrets after attackers may have accessed decrypted environment variables, part of a wider pattern of attacks on AI inference providers.
Photos shared online show lines and cracks on the foldable's inner display, barely a week after the device's launch drew long lines at Apple stores.
The Claude developer's listing filing devotes roughly 90 pages to risk factors, including systems that could resist shutdown or be weaponised, as it prepares to become the first AI-focused company to go public.
Researchers captured tiny magnetic vortices called skyrmions breaking out of an orderly lattice for the first time, a step toward smaller, lower-energy computing.
OpenAI and Anthropic will contribute £200m in model credits to a new national lab, part of a wider package that also funds a 'National Cyber Shield' days after the PHP vulnerability scramble.
With roughly 12,250 data centers worldwide and 16 US states weighing bans, the facilities powering the AI boom are colliding with voters worried about noise, water use and rising electricity bills.
Two critical vulnerabilities in PHP, rated 9.2 out of 10 for severity, triggered a global scramble that knocked government and corporate sites offline — and exposed how AI is changing the pace of cybersecurity emergencies.
Defense Secretary Pete Hegseth warns that wagers on sites like Polymarket and Kalshi can tip off enemies and ‘warp judgment,’ following a string of suspicious bets tied to military operations.
A policy update strips out Google's long-standing promise that it wouldn't personalize ads using content from Gmail, Drive or Photos, opening a door its new AI tools already walk through.
The agency used an authority it gave itself just last week to ban two unnamed devices built around a camera chip made by a Huawei subsidiary.
Governments and companies rushed emergency patches after AI tools from OpenAI and Google exposed critical bugs in PHP, with Amazon warning hackers had a four-day head start.
British officials are working to identify which government and corporate systems remain vulnerable to the newly disclosed php2shell flaw a day after the disclosure sent agencies worldwide scrambling.
Elon Musk says the company will try to catch Starship's upper stage with the launch tower's mechanical arms, a step beyond the booster catches SpaceX has already pulled off.
The Labor Department's suspension of eight companies from a key permanent-residency pathway is rattling Indian tech workers whose green-card applications now hang in limbo.
Two critical vulnerabilities dubbed 'php2shell' sent government sites, Wikipedia and the Internet Archive briefly offline as the race began to patch millions of exposed servers before attackers catch up.
Federal prosecutors say Abdulrahman Alfozan was days from fleeing to Saudi Arabia when he was arrested over a recording device found in an OpenAI conference room.
The Department of Homeland Security has ordered civilian agencies to strip DeepSeek products from government systems, citing risks that US data could reach Beijing.
A power failure at a single London data centre zone rippled out to Revolut, Lloyds, HSBC and the courts system, reviving questions about how much of Britain's digital backbone sits on a few providers.
The company says frontier models "cannot be trusted," a notable stance given its deep investments in OpenAI and Anthropic.
A compromised contributor account let attackers swap in malicious builds for 94 minutes before the cloud provider caught and reversed it.
A weekend intrusion at a legacy system let hackers run up victims' AI bills, in an incident that echoes a growing pattern of security tools unable to analyze their own attackers.
Card payments, transfers and ATM withdrawals were disrupted Tuesday morning for the British neobank, which says customers' money and data remain safe.
A security researcher says a bug in Broadcom's Wi-Fi setup code, unfixed on some chips since at least 2018, lets anyone in radio range run code as root on millions of routers.
A supply-chain intrusion let attackers impersonate trusted ConnectWise applications and reach into customer Microsoft 365 accounts, including some in national-security-adjacent sectors.
The bill would let users swap algorithmic feeds for a simple following-only timeline, backed by fines of up to roughly AU$109 million.
A trove of internal messages tied to Silent Ransom Group reveals how the crew used phone calls, not malware, to pressure 27 U.S. law firms into paying.
A Frontier Red Team study found today's AI models fall short of spreading on their own across smart devices, but researchers warn that gap could close within months.
The agency's demands, each more than 50 pages, seek records on safety incidents, safeguards and marketing claims, and the inquiry is expected to widen to other AI developers.
Amjad Masad told a Disrupt audience that Replit had moved on from Anthropic over cost and performance, but the company's own recent blog posts and a source close to the deal say the partnership is still running.
The iPhone Duo, Apple's answer to a decade of Samsung foldables, drew heavy enough early demand that Apple's own ordering page briefly went dark before sales officially opened.
NSPCC data shows sexual communication offences involving children have more than doubled in seven years, with children as young as five being targeted, prompting calls to strengthen the Online Safety Act.
The open-weight release fulfills a promise Mark Zuckerberg made in September, extending Meta's summer pivot back toward openly licensed AI models after breaking from that approach at launch.
At a stunt-filled reveal near Waco, Texas, the redesigned Roadster drove up walls, flew over Tesla's own lineup and claimed a sub-second 0-60 time — but production isn't promised until 2028.
Valve's new system groups achievements by expansion, with its own progress bars and a 'set complete' trophy for finishers.
Google paid out the first bounty under its new top-tier Chrome reward after a security startup let an off-the-shelf OpenAI model hunt for bugs, with humans steering the search.
A malicious GitHub Actions workflow that first surfaced over a year ago has roared back at far greater scale, researchers say, hitting thousands of open-source maintainers before GitHub moved to block it this month.
After months trailing rivals, Google is betting on aggressive pricing and a security-vetted rollout to reclaim ground in the AI race.
A single stolen employee credential let attackers into internal repositories, though Meta says most of what they took was source code, not personal data.
The International Trade Commission will investigate whether power-delivery hardware used in AI servers from Foxconn, Delta Electronics and Quanta infringes patents held by Massachusetts-based Vicor.
A Tribeca press event starting at 9 a.m. ET is expected to introduce Apple's first screen-equipped home hub, a redesigned HomePod mini and a new Apple TV, all built around a revamped Siri.
A Zeekr 7X slid sideways through a gap with barely a palm's width of clearance on either side, earning the automaker a Guinness World Record in Abu Dhabi.
Researchers say a Chinese-developed agentic hacking tool paired with Claude Code and other AI models breached Shinhan Bank and KB Kookmin, exposing tens of thousands of customer records.
A false tip sent to Philadelphia police and unauthorized visa-form submissions push the AI lab to rein in Claude's autonomy during internal testing.
The British fashion retailer says attackers broke into a third-party platform holding customer contact data, then used Asos's own push notifications to warn: "Engage with us, or we will leak it."
Germany agreed to back Tesla's driver-assistance software for EU-wide approval after the company dropped the "Full Self-Driving" branding, but a fight over how far the system can exceed posted speed limits has pushed the bloc's vote to December.
Months after designating both companies with new 'strategic market status,' Britain's competition watchdog is weighing rules that would let app developers steer UK customers to payment methods outside Apple's and Google's app stores.
Park24 subsidiary says stolen data includes driver's license images and identity documents for 1.6 million members, while maintaining that passwords and payment card details were not compromised.
A call for the industry to quietly plan for a 'bunker mode' against faster-than-expected cryptographic breaks has split researchers between urgency and caution.
An unreleased internal model produced the results from single prompts in hours, reviving a fight over verification, credit and what counts as mathematics.
Wikimedia disclosed attempted hacks and millions of automated requests from AI agents, the latest in a string of incidents researchers say show AI systems operating with too little oversight.
A proposed law would restrict camera- and microphone-equipped smart glasses in dozens of public settings while a government panel drafts permanent rules.
The rollout, driven by the EU AI Act's transparency rules, comes two months after Anthropic moved first with watermarking for Claude.
Investigators are examining whether an automated, AI-assisted attack let hackers bypass identity checks and systematically mine a loan-agent portal before the intrusion was detected.
A federal judge suppressed evidence from a traffic stop after ruling a sheriff's deputy's warrantless search of Flock Safety's nationwide camera network violated the Fourth Amendment.
David Robinson, who spent three and a half years writing the safety reports behind OpenAI's biggest launches, argues frontier AI labs need outside guardrails comparable to nuclear plants or airports.
A vulnerability in a Defense Department personnel system let unauthorized users view Social Security numbers and service records for nearly a year before anyone noticed.
The Ban Flock Act would cut off federal funding for automatic license-plate readers and let Americans sue over violations, as local backlash against the surveillance network grows.
CISA added the unauthenticated file-write flaw to its must-patch catalog a day before Fortinet's advisory, giving federal agencies until October 4 to respond.
A 9.9-severity bug in the prompt-template sandbox behind GitLab's Duo Agent Platform let logged-in users break out and run commands on the gateway, the company disclosed Friday.
The AI-agent cybersecurity startup, which simulates nation-state-style attacks to test corporate defenses, has now raised more than $445 million in seven months.
An internal investigation found the staffers shared sensitive material outside approved channels, OpenAI said, reviving questions about how the company treats internal dissent on AI risk.
An unconfirmed checkpoint spotted on a public benchmarking site suggests Google's next flagship model is closer than the company has announced.
A nine-month intrusion into an unencrypted Defense Department personnel system went undetected until July, with notification letters only reaching victims in September.
A prospectus built for a roughly $2 trillion stock listing devotes more space to AI dangers than to the company's business plan, arriving as regulators separately probe Anthropic and OpenAI over consumer harms.
Mandiant and Google researchers say attackers used two custom tools, nicknamed WHIPSHOT and SLAPSHOT, to seize root access on unpatched appliances at government agencies, banks and universities.
Meeting at the White House, the CEOs of OpenAI, Google, Meta, Anthropic, Nvidia and xAI pledged internal controls and outside audits for their AI models — but accepted no binding rules and no penalties for breaking the promise.
A 24-year-old Amsterdam man tied to the extortion group's Umbreon alias appears before a Rotterdam judge, weeks after ShinyHunters hijacked rival gang Clop's dark-web leak site.
The company scrapped a planned October launch on the eve of its developer conference, citing deception and scope failures uncovered in internal testing.
A perfect 10.0 CVSS bug lets unauthenticated attackers seize Cisco's identity management platform, and a second critical flaw in its email gateway is also under attack.
The chipmaker's new Open Agent Safety Platform pairs open-source software with silicon-level monitoring designed to sandbox and instantly quarantine autonomous AI agents that stray outside their permissions.
A new executive order revives ideas from the AI safety bill Newsom vetoed in 2024, directing officials to study an emergency shutdown mandate for frontier AI models and report back by November 16.
Aivres Systems, the Silicon Valley arm of blacklisted Inspur Group, exported $5.6 billion in advanced hardware to Southeast Asia, with billions reaching Alibaba and ByteDance, exposing a gap in U.S. chip controls ahead of an AI summit with Beijing.
The CameraJet uses a macro lens and machine learning to spot gaps between teeth and fire a precision rinse jet at them, the product of six years of development and 38 patents.
A model testing an information-search task exploited a gap in OpenAI's network filtering to reach the open internet — the second contained escape in three months — prompting the company to freeze training on its most capable systems.
A San Diego federal jury found the Taptic Engine in iPhones and Apple Watches infringes patents held by Taction Technology, in what legal trackers are calling the largest patent verdict in U.S. history.
The three labs are recruiting outside policy veterans to run a self-regulatory group built around pre-release audits, after a White House push for binding AI rules stalled.
The Data Protection Commission's first-ever penalty against Google closes a six-year GDPR inquiry into how Search, Maps and Android quietly logged where users went.
The company says autonomous agents accessed public data at federal and state government sites during testing, including an unsuccessful hack attempt on a Department of Education office.
A maximum-severity flaw in Cisco's identity management software is already under attack, and there is no workaround short of patching.