Recursive Self-Improvement by 2028 - Jack Clark Says It's a Coin Flip Plus Ten

calendar_today Date: MAY 04, 2026

schedule Duration: 1:04

visibility Views: 1

database

Summary Report

Anthropic's Jack Clark puts the odds of recursive self-improvement happening by end of 2028 at 60%. AI systems capable of building their own successors, within roughly two and a half years.

01. Jack Clark, Anthropic co-founder and policy lead, says there's a 60% chance recursive self-improvement happens by end of 2028.
02. Recursive self-improvement (RSI) means AI systems designing and building their own successors with no human in the loop.
03. Clark's forecast is based on weeks of reading hundreds of public AI research sources.
04. The estimate aligns with Anthropic's Frontier Safety Roadmap, which calls fully automated AI R&D plausible by early 2027.
05. Around 70 to 90 percent of code for new Claude models is already written by Claude.

Anthropic co-founder Jack Clark has put a 60% probability on artificial intelligence achieving recursive self-improvement (RSI) by the end of 2028. This forecast comes after weeks of analysing hundreds of public sources on AI development progress. Recursive self-improvement represents a critical milestone where AI systems become capable of designing and building their own successors without human involvement. Clark's prediction carries particular weight given his role as author of the Import AI newsletter and his position at one of the leading AI safety companies. The timeline aligns with other signals from Anthropic. The company's Frontier Safety Roadmap suggests fully automated AI research and development could arrive by early 2027. Jared Kaplan, another Anthropic researcher, recently told Time magazine that such capabilities might emerge within a year. Current evidence supports this trajectory—between 70% and 90% of new Claude code is already written by Claude itself. This 60% probability over roughly two and a half years represents a mainstream view rather than fringe speculation. Given the implications of AI systems that can improve themselves, the focus now shifts from whether this technology will emerge to how society should prepare for its arrival.