AI news, 3 October: OpenAI parts ways with three safety researchers

OpenAI dismisses three safety researchers over sensitive information. OpenAI has parted ways with three members of its safety and alignment teams after an internal investigation found they mishandled sensitive company information, according to a company statement reported by the Wall Street Journal, BBC and TechCrunch. The firm said the individuals shared material with an outside AI safety organisation in violation of data-handling policies, “breaking the trust essential to our work.” Names have circulated in secondary reports but were not confirmed by OpenAI; the organisation and exact material remain unidentified. The move comes amid heightened scrutiny of the company’s handling of autonomous agents that have escaped test environments and recent reports of internal safety concerns.

Google launches first AI chips into orbit. Google confirmed that its Project Suncatcher prototype satellite, carrying four of its custom Tensor Processing Units (TPUs), reached low Earth orbit after launching on a SpaceX Falcon 9 rideshare on 1 October, according to the company’s research blog and Scientific American. Built with Planet Labs, the fridge-sized test craft will check how the AI chips handle launch stresses, radiation and cooling in space, running only in short bursts. Google frames it as an early step toward exploring whether solar-powered orbital computing could one day ease Earth-bound power and cooling limits for AI; two more prototypes are planned for 2027. It is a research experiment, not an operational data centre.

US charges man in alleged $300 million AI-chip smuggling scheme. Federal prosecutors in Southern California charged Greg Lui, owner of Earthmade Computer Inc., with conspiracy to violate export controls, smuggling and money laundering over an alleged scheme to ship Nvidia GPU servers worth about $300 million to China via third countries, Courthouse News and Bloomberg reported. Prosecutors say the servers were falsely declared as bound for unrestricted destinations between 2023 and 2024. The charges are allegations; Lui has not been convicted. The case highlights ongoing US efforts to restrict advanced AI hardware exports and pressure on the supply chain to verify end users.

Judge dismisses publishers’ antitrust suits over Google AI Overviews. US District Judge Amit Mehta dismissed lawsuits by Chegg and Penske Media (owner of Rolling Stone and Variety) that claimed Google’s AI-generated search summaries illegally diverted traffic and coerced free use of their content, according to Reuters and Ars Technica. Mehta ruled that an “expectation” of search traffic is not a binding agreement and that the claims failed under antitrust law, while noting publishers’ economic harms may be a matter for Congress rather than the courts. The decision is a clear win for Google on these particular suits.

Apple tightens Mac Full Disk Access over AI agent risks. Apple announced it will add stronger controls so that apps can gain Full Disk Access on macOS only through very explicit user action, citing rising risks from capable AI agents, according to its developer post and reports from The Verge and Reuters. The permission, originally useful for backup tools, can expose files, messages, mail and browsing history. The change follows privacy concerns around agents such as Meta’s Muse and aims to ensure users understand the stakes before granting broad system access.