Anthropic says AI could soon build its own successors

Anthropic Warns AI Could Soon Build Its Own Successors

Anthropic published a detailed blog post Thursday arguing that artificial intelligence systems are accelerating their own development at a pace that could lead to “recursive self-improvement” — where AI designs and builds its successor with little or no human involvement — sooner than most institutions are prepared for.anthropic

The post, co-authored by Anthropic Institute researcher Marina Favaro and co-founder Jack Clark, draws on internal company data to illustrate the trend. More than 80 percent of the code merged into Anthropic’s production codebase is now authored by Claude, its AI system, up from single digits before February 2025. Engineers at the company are shipping eight times as much code per quarter as they did from 2021 to 2025.anthropic

Internal Data Shows Compounding Gains

Beyond raw code output, Anthropic shared data showing Claude’s ability to handle complex, open-ended tasks has improved sharply. On the most difficult engineering problems — those with no clear specification — Claude’s success rate rose 50 percentage points in six months, reaching 76 percent in May 2026. On a benchmark measuring how well Claude can optimize AI training code, performance jumped from a roughly 3x speedup under Claude Opus 4 in May 2025 to approximately 52x under Claude Mythos Preview in April 2026.anthropic

The company also disclosed that in a March 2026 internal poll of 130 researchers, the median respondent estimated they were producing around four times as much output with AI assistance as they would without it.anthropic

A Call to Build Coordination Mechanisms

“If it were possible to effectively slow the development of this technology to give ourselves more time to deal with its immense implications, we think that would likely be a good thing,” the post states. Anthropic said it would slow down or temporarily pause development if other frontier labs did so in a verifiable manner, but acknowledged that unilateral action would simply shift the lead to a competitor.anthropic

The post arrived one day after OpenAI published its own public policy agenda on Wednesday, which called on the U.S. federal government to “prioritize monitoring progress towards recursive self-improvement” and to strengthen the Center for AI Standards and Innovation as the primary institution for frontier AI safety. OpenAI has also posted job listings for a safety researcher focused specifically on preparing for recursive self-improvement, offering compensation between $295,000 and $445,000.businessinsider

The Road Ahead

Anthropic outlined three possible futures, ranging from capability gains stalling at current levels to full recursive self-improvement where “the pace of progress in AI development becomes determined entirely by the availability of compute”. The company said it plans to organize conversations with policymakers, researchers, and civil society in the coming months to address coordination challenges, comparing the verification problem to Cold War-era arms control treaties — but noting the world does not have decades to build trust this time.anthropic