Altman confirms self-improving systems are running in production

The arc from an operational detail on the 15th to a role on the org chart on the 28th. In a post on X (sama), Sam Altman said OpenAI was running systems that can self-improve and that this called for a Head of Preparedness to manage the recursive ascent. The corpus had followed the same storyline from Codex watching its own training runs through the PostTrainBench leaderboard for models post-training models; this is the first time the lab’s chief executive states it, one day after Roon’s ‘solidly in the takeoff’. The governance half of the remark resurfaces when OpenAI begins monitoring its own coding agents in March.