David Robinson resigned from OpenAI and denounced that "the culture is broken": whoever wrote the safety reports of each major launch is now calling for nuclear-level safeguards. He told about it in an editorial in The Atlantic and was reported by TechCrunch and The Verge on October 3.
Robinson is not a repentant newcomer: he led Safety Systems and was lead writer of the Preparedness Framework v2. His thesis: the problem is not a lack of regulation but of culture – the expertise and processes of a nuclear plant or an airline where error kills are lacking.
You may also likeOpenAI proporciona inteligencia artificial a la OTAN para combatir ciberataquesIt is the second black week in a row: on October 1 the company fired three researchers (Wang, Korbak, Balesni) for alleged leaks, a case that Digitalia covered (note here). And in September Coxon (Anthropic), Benton (Anthropic) and Engels (DeepMind) had already left with the same warning.
Robinson himself admits the cliché: the insider who sounds the alarm after signing the disaster. The Verge picks it up without cynicism: they built this; if they warn, listen.
The specific requirement – nuclear-like processes – points to the real blind spot: laboratories run voluntary audits on systems they have already deployed. The coming regulation (Washington, California, Brussels) will take note of who is asking for what.
What's next: whether OpenAI publishes specific security changes or retains its next frontier model. The exodus of safety is no longer a trickle: it's a hemorrhage with a manifest.
You may also likeEE. UU. limita el acceso remoto a chips de IA desde Singapur, Malasia, Tailandia, Indonesia y BrasilImage credit: Protester with megaphone, Pierre-Selim via Wikimedia Commons, licensed under CC BY-SA 3.0.




