NYT's Klein Urges Regulatory Ban on Recursive AI Self-Improvement

Ezra Klein's argument, in a September 20 New York Times opinion column, is that pacing the frontier is not enough. He wants a comprehensive ban on recursive self-improvement, with firms applying for narrow exemptions rather than working under broad permission.

"If you are losing your ability to evaluate the models you have now, maybe don't let them build models you'll be even less capable of controlling in the future," Klein writes. His line on the industry's preferred "pace the frontier" framing: walking quickly off a cliff, he argues, is only marginally better than sprinting off one.

The column leans on Anthropic CEO Dario Amodei's own words. Amodei has written that recursive self-improvement "is starting to happen across the industry, including at Anthropic, as we and others have described. Left unchecked, it could outrun our ability to understand and control these systems, and so must be pursued very carefully, if at all." Klein reads that caveat as an admission and pushes past it to a rule.

The mechanic in question is straightforward: AI systems doing enough of the work of building the next generation of AI systems that each cycle accelerates. Klein wants the definition drawn expansively because the boundary is grey. Firms who need to keep automating pieces of the pipeline would come to a regulator for a carve-out.

The essay lands the same week rank-and-file researchers at OpenAI, Meta and DeepMind waved off doom warnings inside their own companies, and Politico published its account of the White House's 19-day standoff with Anthropic over a jailbreak.