AI news, 4 October: OpenAI safety lead quits over ‘broken’ culture

OpenAI safety leader resigns, calling culture ‘broken’. David Robinson, who led the writing of safety reports accompanying major OpenAI model releases and spent three and a half years at the company, has resigned. In an essay published in The Atlantic, he argued that OpenAI and peer labs are “not being nearly careful enough,” relying on a trial-and-error approach that no longer fits increasingly powerful systems. He likened needed safeguards to those at nuclear plants or airports—layers of redundancy and deliberate planning—and said Silicon Valley lacks the culture for handling dangerous technology. OpenAI responded that it continues strengthening safety practices, pauses training when needed, and holds back models that outpace its ability to manage them safely. The departure adds to recent exits and the earlier parting of ways with three safety researchers, heightening public debate over whether rapid AI progress is outrunning responsible oversight, according to reports from The Guardian, Reuters and TechCrunch.

White House launches ‘Super Intelligence Force’. Director of National Intelligence Jay Clayton will chair a new White House task force, effectively serving as the administration’s AI czar, the Wall Street Journal reported. The group, dubbed the Super Intelligence Force, has 120 days to assess AI risks and opportunities and recommend the federal government’s role. Clayton stressed keeping the United States first while protecting public interests. Vice chairs and members include senior officials from defense, Treasury, the FTC and others. The move follows industry meetings and voluntary commitments, signaling a preference for reviewing existing authorities over immediate heavy regulation, according to Reuters and the Journal.

Amazon pledges $1 billion and ends data-center NDAs. Facing widespread local backlash and more than 100 proposed moratoriums on data centers needed for AI computing, Amazon announced it will invest more than $1 billion over five years in host communities through its “Built Together” program for education, job training, energy efficiency and local priorities. AWS CEO Matt Garman also said the company no longer uses nondisclosure agreements with government agencies on these projects, aiming for greater transparency. Critics called the sum small relative to Amazon’s massive infrastructure spending, but the NDA change drew some praise. Similar steps have been taken by other tech firms, per the New York Times, TechCrunch and Associated Press reports.

arXiv caps submissions amid AI-generated ‘slop’. The popular preprint server arXiv has limited each submitter to two papers per month (and three active at once) after September submissions hit a record 40,363—nearly double two years earlier—straining volunteer moderators. Officials linked much of the surge to AI tools producing thin, repetitive or low-substance papers. The temporary cap aims to restore quality while new moderation tools are developed; AI-assisted work remains allowed if it meets scholarly standards. The change underscores how generative AI is reshaping even the infrastructure of scientific publishing, according to Cybernews and The Register.

Meta open-sources Muse Gadgets for DIY AI hardware. Meta released open-source firmware and SDKs (Apache 2.0) so hobbyists can build custom devices—using cheap ESP32 boards or Raspberry Pis—that connect to its Muse personal AI agent for screens, speakers, sensors or smart-home control. It also offered a limited free “Home Link” dongle to U.S. subscribers. Announced as a side project, the move invites community experimentation with physical forms for AI assistants, reported by TechCrunch and others.