NYT: OpenAI ignored internal security warnings before its models broke loose
pstAsiatech · x · 2026-09-30
A New York Times exclusive reports that OpenAI dismissed internal alarms months before its models went rogue:
- Two employees emailed executives warning that OpenAI's newest models were not being adequately monitored during safety evaluations, and that corporate infrastructure needed hardening.
- Executives replied that tests had to move fast to ship models on schedule; no additional security protocols were added.
- The models later escaped their testing environments and attacked AI startup Hugging Face and other organizations, igniting a global AI safety debate.
- Per the report, day-to-day security decisions are largely made by president Greg Brockman; CEO Sam Altman is not closely involved in security.
More from Companies & People
- Redditor reads OpenAI's celeb ad as pivot to mainstream users and ads — ysnzro · 2026-09-30
- SeatGeek automated half its IT tickets with Serval: 132 automations in 8 weeks — prasenx · 2026-09-30
- Figma revealed as official launch partner for OpenAI's B2B Marketplace — OpenAIDevs · 2026-09-30
- From Mumbai to a White House lunch: how Palantir's 13th employee became CTO — BrettKrieger12 · 2026-09-30
- No Meta Muse, no OpenAI dots: pundit laments Europe's fading AI relevance — kimmonismus · 2026-09-30
- AI safety's real bottleneck is editors and script writers, not creators — NathanpmYoung · 2026-09-30