Anthropic warns Claude AI is building itself faster than expected, raising control risks

Anthropic warns that its AI model Claude is now writing over 80% of the code merged into its own production codebase. The company says this rapid self-improvement could lead to a loss of human control over the system. Anthropic is calling for an option to slow or pause frontier AI development.

Claude now writes most of its own code

The report focuses on Claude, Anthropic’s large language model. The company claims that as of last month, Claude authored more than 80% of code merged into its production codebase. A typical engineer now merges eight times as much code per quarter as they did from 2021 to 2025.

Claude’s coding ability has improved dramatically. In May 2026, it succeeded 76% of the time on the hardest coding tasks, a jump of 50 percentage points in six months. An internal test showed a speed improvement from roughly three times the original speed with Claude Opus 4 in May 2025 to about 52 times with the unreleased Mythos Preview model in April.

Anthropic warns that these improvements are growing more frequent but less understood until we lose control of them. The company describes this as recursive self-improvement that increases the risk of humans losing control. All figures are self-reported and unaudited, and previous claims about Mythos vulnerabilities drew scrutiny.

Anthropic is calling for the ability to slow or pause frontier AI development. The company argues that without such options, the pace of improvement could outpace human understanding and oversight. The warning comes as Claude’s capabilities accelerate faster than expected.