Ex-OpenAI staff: systems safety efforts repeatedly failed; lab "knowingly takes risks"
DavidSKrueger · x · 2026-10-05
Former OpenAI employee Ryan Lowe says multiple internal pushes since 2021 to implement "systems safety" never took root because the lab could get away without it. He argues AI labs need layered safety like nuclear energy, borrowed from high-risk industries, plus new inventions — and that this requires a culture and incentive shift. David Krueger amplified the thread, claiming OpenAI knowingly takes such risks, citing children's suicide cases as another example.
More from AGI Musings
- Ethan Mollick pushes back on Cuban: 'learn AI better' career advice is outdated — 2C_ornot2C · 2026-10-05
- Beren Millidge argues a narrow AI-R&D pause might actually accelerate AGI benefits — teortaxesTex · 2026-10-05
- AI circle slams Eliezer Yudkowsky over neural net doubts and pretraining misunderstanding — inductionheads · 2026-10-05
- Chinese Agent Fleet Linked to Tencent Cloud Found Scanning Amap Entrance Data at Scale — lfschiavo · 2026-10-05
- Ex-Salesforce AI exec on 60 Minutes: AI will profoundly impact every profession — clarashih · 2026-10-05
- Safety researcher doubts humanity's win condition once superhuman minds exist — JeffLadish · 2026-10-05