Safety Transparency Lead, Safety Systems · OpenAI · 2026
Evidence reviewed
Safety-transparency leader who resigned from OpenAI in late September 2026. Robinson said the company's sprint from one launch to the next prevented it from achieving the care needed for increasingly capable systems. He argued that OpenAI's trial-and-error approach guarantees periodic failures, called for safety practices modeled on nuclear power and aviation, and said he could strengthen incentives for safety more effectively from outside the company.
OpenAI confirmed that Robinson left in the week before October 3, 2026. Robinson described the timing as “this week” in an essay published October 3. This establishes a late-September or early-October departure, but the exact day is not publicly documented; September follows the company-confirmed reporting.
“As the company sprints from one launch to the next, it is failing to achieve the level of care that I believe is needed.”
Robinson spent three and a half years at OpenAI, first in public policy and later in Safety Systems. He led the drafting of the current Preparedness Framework and oversaw safety reports for 12 frontier launches. He said he valued his colleagues and believed AI could be useful, but concluded that the company's culture, pace, and reliance on iterative deployment were inadequate for systems that might become more capable than humans. OpenAI said it was strengthening security, third-party evaluation, and real-time monitoring, and would pause training or hold back models when necessary.
A safety assessment of OpenAI o1 covering model behavior, evaluations, external testing, and Preparedness Framework risk categories. Robinson is listed among the authors.
OpenAI's framework for tracking frontier capabilities that could create severe harm, defining capability thresholds, safeguards, evaluations, governance, and disclosure commitments. Robinson said he served as lead drafter for this version.
Robinson explains why he resigned from OpenAI. He argues that the company's launch-driven culture and reliance on iterative deployment cannot provide the care required for increasingly capable systems, and calls for operational safety practices drawn from aviation and nuclear power as well as new alignment science.
0 open forecasts · 1 warning and contemporaneous claims
Trial-and-error deployment will produce larger safety failures
“But this approach, by its very nature, guarantees periodic failures—and the scale of those failures is growing as systems get more capable.”
Get an email when a new, source-verified profile is published.
Confirm your address by email. You can unsubscribe at any time, and we never share your address with third parties.