Jan Leike

Alignment researcher; former co-lead of OpenAI's Superalignment team (resigned May 2024); now leads alignment at Anthropic

Anthropic (formerly OpenAI)

Safety researchers

Leike ran OpenAI's team tasked with controlling superhuman AI and quit in 2024 saying safety had 'taken a backseat to shiny products'. He now leads alignment work at Anthropic. He has agreed with a 10-90% range for catastrophe—essentially saying the honest answer is deep uncertainty.

Their number

10–90%

When host Rob Wiblin said his own p(doom) was 'more than 10%, less than 90%', Leike replied: "Yeah, I think that's probably the range I would give too." (80,000 Hours Podcast, August 7, 2023) — note this is agreement with an interviewer's range, not an independently volunteered figure.
Source

On timing

Superintelligence "could arrive this decade" (OpenAI Superalignment framing, 2023); "systems could get a lot smarter or a lot more capable over the next few years."

Source

What they call for

  • Labs to devote serious compute and priority to alignment
  • Safety culture over product speed
  • Scalable oversight and automated alignment research

In their own words

  • "But over the past years, safety culture and processes have taken a backseat to shiny products." (resignation thread)

    17 May 2024

    Source

  • "I have been disagreeing with OpenAI leadership about the company's core priorities for quite some time, until we finally reached a breaking point."

    17 May 2024

    Source

Signed

  • CAIS extinction statement 2023 (reported; not independently re-verified on this pass)

Last verified 1 September 2026

Report a correction

Back to all voices