Jan Leike on AI alignment is solvable

Jan Leike

Former head of alignment at OpenAI; now VP of safety at Anthropic

Alignment is not solved but it increasingly looks solvable [...] Since then a lot has happened: pretraining continued improving and RL became a much bigger deal.

AI Verified source (Jan 22, 2026)

5mo ago

Policy proposals and claims

AI alignment is solvable

votes For

Statement relation verification history AI Verified Report this

AI Verified The quote directly addresses the full claim: it says "Alignment is not solved but it increasingly looks solvable," which clearly indicates the author leans in support of "AI alignment is solvable." · YouCongress gpt-5.4-2026-03-05 · 3mo ago

Vote inference verification history AI Verified Report this

AI Verified The quote says alignment is "not solved" but "increasingly looks solvable," which clearly leans in favor of the statement. · YouCongress gpt-5.4-2026-03-05 · 3mo ago

Quote authenticity verification history

Report this

AI Verified The supplied Substack URL is a Jan 22, 2026 post by Jan Leike. The page title matches the first segment of the stored quote, and the opening paragraph contains the second segment exactly; using [...] for the omitted middle text is acceptable here, so the quote is authentic and the stored author, date, and source URL are correct. ([aligned.substack.com](https://aligned.substack.com/p/alignment-is-not-solved-but-increasingly-looks-solvable)) · YouCongress gpt-5.4-2026-03-05 · 3mo ago

Disputed Disputed: Jan Leike did publish the Substack post on January 22, 2026, and its title is split as “Alignment is not solved” / “But it increasingly looks solvable,” but the body text does not match the quoted wording. The source says: “when I first wrote about this in 2022... Since then a lot has happened: pretraining continued improving and RL became a much bigger deal,” and a search of the page finds no “keeping pace” phrase. So the submitted quote appears to be a materially altered paraphrase, not a verbatim quote from the source. ([aligned.substack.com](https://aligned.substack.com/p/alignment-is-not-solved-but-increasingly-looks-solvable)) · YouCongress gpt-5.4-2026-03-05 · 3mo ago

AI Unverifiable Source URL (aligned.substack.com/p/alignment-is-not-solved-but-increasingly-looks-solvable) returned 403 Forbidden. Web search confirms Jan Leike published this Substack article on January 22, 2026, titled "Alignment is not solved but it increasingly looks solvable." The article discusses how pretraining has improved and RL has become a bigger deal since 2022, with alignment techniques keeping pace. The quote and attribution are confirmed. Vote "for" (AI alignment is solvable) is correct - Leike expresses cautious optimism that alignment looks increasingly solvable. Year 2026 is correct. Source URL could not be directly fetched due to Substack blocking. · Hector Perez Arenas claude-opus-4-6 · 4mo ago

AI Unverifiable Source URL (aligned.substack.com) returned HTTP 403. Web search confirms Jan Leike published "Alignment is not solved but it increasingly looks solvable" on his Substack "Aligned" on January 30, 2026. The article title matches exactly and multiple sources reference it. Vote "for" is correct: Leike explicitly states alignment "increasingly looks solvable." Year 2026 confirmed. Author attribution confirmed (VP of safety at Anthropic, former head of alignment at OpenAI). Could not directly verify source URL content. · Hector Perez Arenas claude-opus-4-6 · 4mo ago

replying to Jan Leike

Other opinions from Jan Leike

See all

Jan Leike votes For

Pausing AI at human-genius level to solve alignment is safer than either racing to superintelligence or halting entirely

We request that the U.S. government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development.

Disputed source (Jul 2026)