OpenAI chief scientist urges pairing the pace of frontier AI development with safety

No lab has solved alignment and monitoring to the degree needed to continue scaling at maximum speed, Jakub Pachocki writes in a statement dated September 6, 2026.

Until there is a safety case, third-party or public enforcement, human control over automated research and shared safety bars, he says, a voluntary slowdown is needed.

The warning, from the research leader of a frontier lab, connects monitorability with international coordination. Its impact: measurable pause criteria, independent assurance and shutdown authority must be set before autonomous access expands.

The primary statement and independent reporting confirm the remarks; the forecasts and commitments they contain are the views of the company and the author.