OpenAI has removed every fourth production engineer from new tasks to strengthen its model protection. Greg Brockman on the podcast doesn't. Priors called it a "security freeze": while it lasts, these engineers don't release new features and focus only on security. The whole story takes half a minute, from the 31st minute. A quarter of engineers who defend against their model talk about the risks more persuasively than any other statement. If your AI agents have access to code and servers, check their permissions before the next release: what they can run themselves and who's reviewing it.

OpenAI has removed every fourth production engineer from new tasks to strengthen its model protection. Greg Brockman on the podcast doesn't. Priors called it a "security freeze": while it lasts, these engineers don't release new features and focus only on security. The whole story takes half a minute, from the 31st minute. A quarter of engineers who defend against their model talk about the risks more persuasively than any other statement. If your AI agents have access to code and servers, check their permissions before the next release: what they can run themselves and who's reviewing it.