David Sacks calls recursive self-improvement the most dangerous AI work and says Anthropic is racing toward it
All-In Podcast · Video interview
Read the source (opens in a new tab)- What happened
- On the All-In podcast posted November 13, 2026, co-host David Sacks said "recursive self-improvement is the most dangerous thing anybody can be working on."
- Key points
- He said Anthropic is "saying that recursive self-improvement has kicked in despite the obvious safety problems with that" and is "treating it like a business advantage."
- He said Anthropic is "doing more recursive self-improvement than anybody else" and called its blog post about pacing the frontier "lip service."
- He said "if you think that this type of development is risky, stop doing it. But they have not stopped," adding that Anthropic is now marketing it in its IPO road show.
- He said he has "never quite understood why all the safety groups love them so much," while adding "I'm not saying they're doing things in the riskiest way."
- Evidence
- Podcast recording, an argument by a co-host who also praises Anthropic as an investor. Quotes come from YouTube's automatic captions; the passage is Sacks's answer after the host asks him by name at 1:14:50.
- Uncertain
- What Anthropic says about recursive self-improvement and in its road show is Sacks's account; Anthropic's response is not part of this source.