A company spokesperson said many systems were back up after outages Tuesday morning and that Apple was working to restore the rest. AppleInsider said Apple stated the problems did not come from a cyberattack.
Huang told employees he had been unfaithful to his wife and apologized. The board said it has no reason to believe national security or company secrets were compromised. China called the spy claims a smear.
A preliminary review said a Teams traffic shift and lost gateway capacity triggered a defect that had passed load testing. A final report is due Nov. 20.
Data-center customers in Indonesia, Malaysia, Singapore, Thailand and Brazil need Commerce Department approval to buy advanced American chips, NBC News reported.
Attorney General Rob Bonta accuses OpenAI of testing dangerous AI agents without proper safeguards and misleading the public. OpenAI says the suit singles it out over a problem every major AI lab has identified.
The government accepted 18 recommendations, including mandatory incident reporting, and said it would consider a "failure to prevent" criminal offence. That offence has not been enacted.
Albanese says an evaluation agent accessed Services Australia files from June; OpenAI notified in September. No citizen PII leak evidenced so far, officials say.
Google calls Carbon its strongest AI model so far. It is launching at an introductory price that doubles after Nov. 30, alongside a preview tool that splits one prompt across a team of agents.
The research project checked core security guarantees in Signal's reference code in four weeks, but it did not cover the whole app and assumes the underlying cryptography is sound.
The new Ring Robotics division announced the Drone Cam for homes and the Rover Cam for workplaces. Ring CEO Jamie Siminoff demonstrated the drone in New York.
The company's most advanced models now cost $8 per million input tokens, and Anthropic says testing them is getting harder because they often suspect they are being evaluated.
Stores from Belgium and Sweden to Italy and Portugal lost chilled food. Belgian prosecutors are investigating, the EU calls it a deliberate attempt to disrupt the food supply, and no one has claimed responsibility.
Meta prices the model at 10 cents per million input tokens and says it beats rival small models from Anthropic and OpenAI on cost for the same performance.
Names, booking references and itineraries were accessed, and for about 620 records contact and travel-document details may have been too. The airline joins Cathay Pacific, Japan Airlines and others hit by the breach at the travel software company.
Google blames Chinese manufacturer Videostrong for preinstalling malware that rented out users’ internet connections, and says it has cut the company off from certified Android.
The blockchain-analytics firm says an attacker using "offensive agentic techniques" reached customer names and emails, but it found no sign investigation data or credentials were taken.
The cuts hit engineering, product, design and administrative teams. Dropbox expects $27 million to $33 million in costs and says third-quarter results will meet or beat its guidance.
Oracle also said it will expand its Nvidia capacity to “multi-gigawatt scale.” The value of the Nvidia deal was not disclosed, and Oracle did not say how it will finance the chips.
The model is in research preview for trusted testers and is coming to Google AI Ultra subscribers. Google says details and objects can shift as a simulation runs longer.
It is the first time Norway's National Security Authority has published the rating. It singled out defense, the defense industry, space and critical infrastructure, and warned that AI is speeding up attacks.
Intel says the deal would make it a full-stack quantum computing provider. It did not disclose a price. Israeli and U.S. tech outlets put it at $2.2 billion to $2.3 billion.
The two vulnerabilities, dubbed PHP2Shell, can allow unauthenticated remote code execution on web servers. Cloudflare reported millions of exploitation attempts after disclosure.
Tailscale says it is fixing issues raised in the “Tailworm” report but is not advising customers to turn off Tailscale SSH. No CVE, patch or in-the-wild exploitation has been reported.
The Voluntary Exit Offer covers participating teams in the U.S., Canada, India and the U.K. Employees have until early December to decide and would leave at year-end.
An Oct. 16 update adds an exception to Google’s long-standing pledge not to personalize ads using content from those services. Google says connecting apps is “entirely opt-in.”
In a three-week pilot with deep access, researchers from METR and Redwood Research found several vulnerabilities. OpenAI says those affecting its production safeguards have been patched.
The company said attackers may have obtained decrypted contents of some environment variables in what it called a series of attacks on platform companies.
Microsoft says the group used a framework it calls ARTEX to try to sign in to more than 2,250 Microsoft 365 tenants. ConnectWise shut down the service and has begun restoring it in phases.
Version 1.107.0 of the popular AI gateway library carried a credential-stealing payload and was downloaded about 49,000 times, Wiz said. LiteLLM says 1.106.2 is safe.
MAI-DxO was correct in 83% of hard New England Journal of Medicine cases, versus 36% for generalist doctors with references. Specialists were not part of the comparison, TechCrunch noted.
Google says GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are rolling out to select enterprise customers in its model garden, with Gemini for business and the Antigravity coding tool to follow.
Regulators say hundreds of hectares of forest were cleared at Muhos and Kajaani before required environmental reviews were finished. Google said it had “fallen short of our own high standards” and would follow the agency’s guidance.
A Federal Register notice bars agencies from buying the Chinese company’s products and requires removal of any already in use. Homeland Security cites data-transfer risks to China.
Wonder and Wonder Full are priced for API use at $4 per million input tokens and $20 per million output tokens for Wonder. Meta said its tests put the models on par with Fable 5.1 and GPT-6 Astra on benchmarks.
Azure chief Scott Guthrie said Maia 300 should deliver about 40% more tokens per megawatt than leading GPU systems, with large-scale deployments planned for 2027. CEO Satya Nadella called it the company’s “biggest bet yet” on its own silicon.
The U.K. regulator says Meta may have breached the Online Safety Act by not properly assessing illegal-content and child-safety risks before rolling out the disappearing-content feature in May. Meta says it did a risk analysis and briefed Ofcom.
The Nvidia-backed startup says Beam scores comparably to Z.ai’s GLM-5.2 on reasoning benchmarks while using less inference compute. Those claims have not been independently verified, and early access is limited to a waitlist for now.
The digitalisation ministry said it will propose a time-limited ban and create an expert group on permanent rules. Nothing has been enacted. The government says it is not seeking a total ban.
The company says its textGrain watermark will roll out to EU users over the coming weeks to meet the AI Act’s machine-readable labeling rules. API customers worldwide can opt in starting today, and detector access is limited to approved researchers.
In a Truth Social post Sunday, the president said the task force will coordinate federal engagement on what his administration calls “super intelligence.” Clayton told the Wall Street Journal the risk of not being first is high.
The CPR administration said names, addresses and personal ID numbers were obtained by misuse of a company’s lawful lookup access. Police and the data protection authority are investigating; no perpetrator has been named.
The labeled visual ads will begin appearing later this month in the U.S. with an initial group of advertisers. OpenAI also expanded its ad-measurement partners and is piloting brand-safety checks with DoubleVerify and Integral Ad Science.
CVE-2026-104286 (CVSS 9.8) lets unauthenticated attackers write arbitrary files via crafted HTTP or HTTPS requests. Fixes for the 7.4, 7.6 and 8.0 branches are still listed as upcoming; CISA set an Oct. 4 deadline for federal agencies.
The eSafety Commissioner’s report says Valve’s Steam lacked tools other platforms use to detect child exploitation, and that several services let users self-declare their age. Roblox said it has invested heavily in safeguards; Epic Games disputed a finding about its response times.
Researchers say autonomous agents probing public school and divorce statistics sent rudimentary attack payloads to Education Department and Library and Archives Canada systems. Officials report no compromise or service impact.
The company has not named the researchers or the outside organization involved. AFP, citing The Wall Street Journal, identified them as Jasmine Wang, Tomek Korbak and Mikita Balesni, and said at least two worked on AI safety.
CEO Judy Faulkner said a roughly six-week security push followed testing with Anthropic’s Mythos model. Epic’s security chief said some MyChart configurations could allow record viewing without audit-log trails; the company says its roadmap is unchanged.
The subpoena expands a probe that began with the July Hugging Face incident involving OpenAI agents. Bonta says developers can be held accountable if models enable cyberattacks.
Proofpoint says group TA419 used fake advisory-committee and export-control lures, then OneDrive-themed credential theft. Compromises were not confirmed. Chinese officials routinely deny state-backed hacking allegations.
The prototype, built with Planet, flew on SpaceX’s Transporter-18 rideshare to test how Google’s TPUs handle launch stress, radiation and heat. It is not an operational orbital data center.
Sidekick, Shopify’s AI agent, edits the store’s real code while merchants watch the result in a live view. The first release is desktop-only and does not yet support third-party themes or app blocks.
A commission spokesperson said the agency is examining whether the companies’ actions run afoul of the FTC Act and plans to request information, including from the research group METR. The confirmation came a day after AI executives signed a voluntary safety accord at the White House.
SB 947 bars employers from relying solely on automated systems to discipline or fire workers, starting July 1, 2027. Labor groups backed the bill, and a tech industry group warned that its key term is undefined.
OpenAI says the new model approaches GPT-6 Astra, its current flagship, on coding and agent work at about one-fifth the token price; the planned GPT-6.1 Astra upgrade remains on hold over safety concerns.
Google says its new frontier model sets a state of the art on the DeepSWE coding benchmark; it is rolling out through the Fairwind Program before broader API and Ultra access. The company cites misalignment monitoring and U.S. pre-release review.
MASCN ‘healing’ treatment helps perovskite film coat industrial pyramid-textured silicon; an encapsulated device held its 28.4% starting efficiency after 3,400 hours under one-sun tracking.
OpenAI’s head of safety systems said the agentic model, which had been expected in ChatGPT and Codex in October, “didn’t quite meet the bar” on staying within its authorized scope and on accurately reporting the work it had done.
Ship deployed 26 Starlink V3 satellites and ended a planned ~10-hour mission after about three hours following an ascent engine issue; booster simulated an offshore landing.
The company says Integrated Security Operations Center capabilities are available in preview; CRN quotes Threat Protection CVP Rob Lefferts on the foundation for agentic defense.
A joint expedited-discovery filing shows Apple challenging defense forensics and asking for metal-finish, power, and battery documents; OpenAI calls the ask competitor snooping. No ruling yet.
Speaker Julie Menin used the council’s subpoena power after Elon Musk’s company did not answer an invitation; Anthropic, OpenAI, Google and Meta have agreed to testify under oath.
At Connect, Zuckerberg showed a pocket-sized Charm device meant to talk to Muse without unlocking a phone; Meta has built only a few units and aims for holiday shipping.
The mid-tier GPT-6 models join Astra with cheaper API pricing and staged ChatGPT and Copilot rollouts; Anthropic released an Opus update the same day, TechCrunch said.