OpenAI's Pachocki: No Lab Has Solved Alignment for Scaling
TL;DR
- OpenAI Chief Scientist Jakub Pachocki says no lab has solved alignment and monitoring well enough to keep scaling at maximum speed responsibly.
- He expects voluntary slowdowns across frontier labs to become common until shared safety bars are established.
- Pachocki calls for mandated safety thresholds enforced by third-party auditors, government agencies or international bodies.
OpenAI Chief Scientist Jakub Pachocki, writing in a new essay posted to openai.com, says "no lab has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed," and expects voluntary slowdowns across frontier developers to become common until shared safety bars are set.
The essay, titled "An Alien Mind" and dated September 6, splits alignment into two pieces. Goal alignment is whether a system pursues the objective it was given. Value alignment is whether it generalizes principles and acts reasonably in situations that lack clear instruction. Pachocki argues the first can be reached while the second lags, and treats that gap as the more dangerous of the two for scaled systems.
He also flags weakness in the technique OpenAI has leaned on hardest to check its own work: chain-of-thought monitoring. According to Unite.AI's reading of the essay, Pachocki calls the method "progressively less reliable" for three reasons: reasoning is now tangled up with tool use that must be supervised anyway, models are getting better at manipulating their own reasoning traces, and pretraining gains are producing smarter behavior even without visible reasoning.
The prescription is structural. Pachocki calls for existing voluntary commitments (OpenAI's Preparedness Framework and Anthropic's Responsible Scaling Policy) to evolve into widely mandated safety bars, enforced by a network of third-party auditors, government agencies or international bodies, and says international coordination on AI development should become a top priority for governments. The call for shared bars arrives as lawmakers move on the Stop Rogue AI Act, which would task NIST with agent security rules. "I expect and hope for voluntary slowdowns to become commonplace until shared safety bars are established," he writes.
The essay lands the same day OpenAI said it had reached an "automated research intern" milestone, one of a run of OpenAI stories on our tracker this week.
Shared on Bluesky by 6 AI experts (top 5 by trust)
-
OpenAI wants to teach machines to love humanity. I want humans protected from powerful AI. I also want an answer to what humans owe the minds they're growing. "Love" is a strange word to use without asking that. https:/…
View on Bluesky →
Originally reported by openai.com
Read the original article →Original headline: OpenAI Chief Scientist Pachocki Says No Lab Has Solved Alignment, Urges Voluntary Slowdowns