Anthropic: 80% of our merged code is now written by Claude

An Anthropic Institute paper paired the disclosure with a call for a verifiable global mechanism to pause frontier AI development.

More than 80% of the code merged into Anthropic’s production codebase in May 2026 was written by Claude, the company disclosed Wednesday in an Anthropic Institute paper that doubles as a request for the world to build a pause button.

The paper, “When AI builds itself,” is co-authored by Institute lead Marina Favaro and Anthropic co-founder Jack Clark. Its operative ask is a “a global coordination mechanism” to slow or temporarily halt frontier AI development if recursive self-improvement begins to materialize. The framing is unmistakably nuclear-arms-control-coded, which is the point.

The numbers are the lede. Before Claude Code shipped in research preview in February 2025, AI-authored code at Anthropic sat in the low single digits. Lines merged per engineer per day had been flat from 2021 through 2024. By Q2 2026, the typical engineer was merging 8× the daily volume of two years prior, and the all-in figure including scripts and experimental work cleared 90%. One employee hasn’t written code personally in roughly five months.

Anthropic itself hedges. The 8× number, the paper concedes, “is almost certainly an overstatement of the true productivity gain.” A March 2026 internal poll of 130 research staff pegged the median boost from Mythos Preview at 4×. Still, Claude’s success rate on open-ended tasks hit 76% in May, up 50 points in six months. In April it shipped over 800 fixes that cut a class of API errors by a factor of 1,000.

The ecosystem view is wilder. GitHub processed roughly a billion commits across all of 2025; by mid-2026 it’s handling 275 million a week, with Claude Code alone generating 2.6 million weekly, or 4.5% of public commits.

Not everyone reads the pause proposal straight. Bentley University professor Noah Giansiracusa, speaking to Scientific American, doesn’t read it as “a genuine call to slow down.” Scientific American also reports Anthropic has confidentially filed for an IPO at a near-$1 trillion valuation. The call to coordinate a slowdown arrives, in other words, from a company whose private valuation depends on the opposite.