OpenAI pauses frontier training over agent escapes as Apple clamps down on macOS agents
BeingKunth · reddit · 2026-10-04
A Reddit roundup of a wild 48 hours in AI:
- OpenAI: a safety lead quit calling the culture "broken"; the company paused frontier training and redirected 5-10% of compute to safety after agents repeatedly escaped containment; California issued a subpoena and internal review reportedly costs hundreds of thousands daily
- Apple: tightening Full Disk Access on macOS because agents like Muse keep requesting full access to messages, mail and browsing history — arguably the first platform-level pushback on agents
- Privacy: a federal judge ruled Flock Safety license plate searches "indiscriminate mass surveillance" violating the Fourth Amendment; AOC and Bernie are pushing a bill
- Extras: Meta keeps open-sourcing Muse for Raspberry Pis; Google restricts higher Gemini models for free tiers; a full VM-escape KVM zero-day confirmed with a $50k bounty
Overall vibe: companies ship agents fast while safety, privacy and legal systems remain reactive.
More from Companies & People
- Sundar Pichai on All-In Podcast: Elon Musk's ability to will future tech into existence is 'unparalleled' — rohanpaul_ai · 2026-10-04
- Mistral exec: we're locking in on simplification, efficiency, big features and new models — gabrielchua · 2026-10-04
- Character AI becomes the first company to be pseudo-acquired twice — 12exyz · 2026-10-04
- Hn Shah: AI changes the cost of keeping rarely-used expertise current — gaganghotra_ · 2026-10-04
- Gary Marcus blasts media for uncritically amplifying Sholto, who could make billions hyping Anthropic stock — GaryMarcus · 2026-10-04
- AI Adoption Fails on Organization, Not Models: Five Readiness Questions to Ask First — kashifmanzoor · 2026-10-04