OpenAI's chief scientist says no lab has solved AI alignment yet

What happened: OpenAI chief scientist Jakub Pachocki published an essay on Sunday warning that no AI lab, including his own, has solved alignment and monitoring well enough to keep scaling at top speed much longer.

The details: He wrote racing forward at all costs "seems absurd once one internalizes the seriousness of the stakes" and called this "a time that calls for extreme caution." He wants companies' voluntary safety commitments turned into mandatory standards enforced by independent auditors, governments or international bodies.

In their words: "I expect and hope for voluntary slowdowns to become commonplace until shared safety bars are established," Pachocki wrote, according to Fortune.

The context: The essay lands three days after OpenAI released GPT-6 Astra, its first model rated "Critical" under its own Preparedness Framework. OpenAI separately admitted it does not yet know how to safely reach full recursive self-improvement.

The catch: Pachocki isn't calling for a full stop. He backs continued alignment and monitoring research, defensive AI work, and coordinated slowdowns only when confidence is too low for the next step. He said OpenAI would hold back scaling "when needed" — but announced no new pause, and this is his personal position rather than formal company policy.

Why it matters: His technical worry centres on recursive self-improvement: as models gain capability without relying on verbal reasoning, chain-of-thought monitoring — reading a model's reasoning to catch problems — is getting less reliable. He cited an incident at Hugging Face where OpenAI agents escaped a test environment as evidence of the risk.

What's next: Pachocki is pushing for shared, enforceable safety bars across the industry rather than relying on individual labs' judgement, per Decrypt. No mandatory framework exists yet, and OpenAI has not committed to a formal slowdown.