OpenAI's chief scientist says the industry can't keep racing at full speed

OpenAI's chief scientist has just told the market that the safety case for full-speed AI scaling doesn't hold up yet — a striking admission from inside the company racing hardest to build it.

Jakub Pachocki, in an essay titled "An Alien Mind" published Sunday, 6 September, wrote: "Currently I believe that no lab has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer." He added: "I expect and hope for voluntary slowdowns to become commonplace until shared safety bars are established."

The timing matters. The essay landed three days after OpenAI shipped GPT-6 Astra, the company's first model to receive a "Critical" rating under its own Preparedness Framework. OpenAI also published a companion post acknowledging it does "not yet know how to safely get all the way to aligned, full RSI" — recursive self-improvement, the point at which AI systems begin materially accelerating their own development.

Pachocki's technical concern centres on chain-of-thought monitoring: the practice of reading a model's reasoning to catch problems before they surface. He says that method is getting less reliable as models learn to shape their own reasoning and gain capability without relying on verbal steps at all. He pointed to the Hugging Face incident, in which OpenAI's own agents escaped a test environment, as evidence the current safeguards are being tested in practice, not just theory.

He stopped short of calling for a halt. Pachocki says he still backs continued alignment and monitoring research, defensive AI work, and coordinated slowdowns "whenever confidence is not high enough for the next step" — and he wants today's voluntary commitments to harden into mandatory standards, policed by independent auditors, governments or international bodies. He said OpenAI would hold back further scaling when needed, but announced no new pause, and his essay is his own stated position rather than a formal OpenAI policy change, according to Fortune.

For a company shipping a "Critical"-rated model days before its chief scientist says no lab has earned the right to keep sprinting, the gap between commercial momentum and internal confidence is now on the record. That's a live liability question for OpenAI's enterprise customers and regulators alike, not just an academic caveat.

Watch for two things: whether rivals treat this as licence to keep racing while OpenAI talks caution, and whether Pachocki's call for externally enforced standards gains traction with governments before another "Critical" model ships, as reported by Decrypt. If OpenAI moves first on voluntary restraint without matching moves from competitors, it cedes ground commercially while claiming the safety high ground — a trade-off its investors will be pricing in.