OpenAI Scraps Frontier AI Model After Internal Tests Fail, Raising 99% Safety Proof Questions
Updated
Updated · The Guardian · Oct 6
OpenAI Scraps Frontier AI Model After Internal Tests Fail, Raising 99% Safety Proof Questions
3 articles · Updated · The Guardian · Oct 6
Summary
OpenAI withdrew a new frontier AI model after internal safety tests flagged concerns, halting its release before launch.
The decision has renewed debate over whether regulation alone can govern advanced AI when developers have not published detailed safety cases or independently assessable risk analyses.
Existing safety-critical industries such as aviation and nuclear power rely on formal safety cases that typically must show at least 99% confidence that catastrophic accidents are extraordinarily rare.
That benchmark is already hard to meet for systems that could kill about 1,000 people in a single accident, underscoring how much harder it would be to prove safety for frontier AI with far larger claimed risks.