Evan Hubinger: Anthropic has no plan yet to align superintelligence

Anthropic’s Alignment Science lead Evan Hubinger replied on X that he agrees AI could kill all humans, puts his personal estimate above 10% within the next decade, and says the lab does not yet have a plan to align superintelligence.

A named Alignment Science lead saying the lab does not yet have a plan for superintelligence — and is not clearly on track — is a public statement you can quote. The percentage stays his opinion, not a measured event.

The reply sat on a public resignation. On or around 9 September 2026, Anthropic researcher Jacob Coxon left the lab and warned that frontier companies are racing toward self-improving AI and “gambling with our lives.” That is the Coxon event. This filing is Hubinger’s.

Hubinger wrote: “Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade.” He added: “I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.” Those are his words on X, not a lab press release.

The “>10%” line is Hubinger’s personal estimate — more than a one-in-ten chance, in his view, that AI could kill all humans within ten years. It is not a measurement, a company forecast, or a dated incident. CNBC and Forbes quoted the same post the same day.

The prediction market Kalshi amplified the exchange on X with the paraphrase “Another Anthropic safety lead warns ‘humans may not survive the AI race.’” Treat that as radar. The record is Hubinger’s wording, not Kalshi’s headline.