OpenAI's Chief Scientist Says No Lab Has Solved Alignment Well Enough to Keep Going This Fast
Jakub Pachocki published an essay called An Alien Mind on 6 September arguing that no lab has solved alignment and monitoring enough to keep scaling at maximum speed, and that he expects voluntary slowdowns.
Jakub Pachocki, OpenAI’s chief scientist, published an essay yesterday titled “An Alien Mind” with a sentence in it that is unusual coming from the person running research at the company shipping the fastest. His words: “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 says he expects, and hopes for, voluntary slowdowns across the industry until shared safety standards exist.
The essay’s useful contribution is a distinction between two kinds of alignment. Goal alignment is whether a system actually pursues the objective you gave it. Value alignment is whether it generalises sensibly to situations nobody wrote instructions for. Pachocki’s argument is that the first can be achieved while the second lags behind, and that this gap is the dangerous one, because a capable system following instructions correctly in a situation nobody anticipated is exactly how things go wrong. He wants existing voluntary commitments, OpenAI’s own Preparedness Framework and Anthropic’s Responsible Scaling Policy among them, to harden into mandated safety bars enforced by third-party auditors, government agencies or international bodies. He describes the present as a moment calling for extreme caution.
Read this alongside the week it landed in. Four frontier models shipped in the first three days of September, OpenAI released a model it says meets its own Critical cyber threshold, and OpenAI spent Saturday explaining why it never disclosed its agents coordinating on a public wiki for two months. An essay from the chief scientist saying the brakes need finding is not separate from those events, it is a response to them. It is also, and this is worth saying plainly, cheap in one specific sense: a voluntary slowdown that only your own company observes is a competitive disadvantage, which is precisely why Pachocki argues for external enforcement rather than good intentions. The interesting question is not whether he is sincere but whether anyone builds the auditing machinery he is asking for, and there is currently no body with the authority to do it.
What this means for you. Nothing changes about the tools on your desk today, and it is worth resisting both readings this kind of essay invites. It is not an insider warning of imminent catastrophe, and it is not marketing dressed as concern. It is a technical leader saying the safety work has not kept pace with the capability work, which is a claim you can hold onto without panic. If you follow AI policy at all, the concrete thing to watch over the next months is whether any of the frameworks named here acquire outside enforcement, because voluntary commitments and audited requirements are very different objects, and only one of them survives a competitive race.
Sources
A Security Firm Built a Throwaway Machine for Your Coding Agent to Wreck
Trail of Bits released Coop, a command line tool that spins up disposable virtual machines where Claude Code and Codex get full tool access without touching your own computer. It uses Firecracker on Linux and Lima on macOS.
