Robinson managed safety reports for 12 frontier launches and developed system cards to document model behavior before his exit. In an essay for The Atlantic, he characterized OpenAI's safety culture as broken and argued that AI labs should adopt the high-reliability engineering and layered redundancies used in nuclear power plants and aviation. His departure follows the dismissal of three researchers for allegedly leaking sensitive data and other high-level resignations from the company's safety team.
Wednesday, October 21, 2026
Story
OpenAI Agentic Compute Use Jumped 100x per Departing Safety Lead
AI8 days ago
A former safety leader at OpenAI believes the company is building artificial intelligence far more capable than humans are prepared to align or control. David Robinson told Ezra Klein that researchers at OpenAI utilized more than 100 times more agentic compute per day in September than nine months prior, signaling a rapid increase in capability. Robinson resigned after concluding the firm was on track to create systems that could 'think circles around' people while remaining unprepared at a safety and alignment level.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.