Latest
  1. Johns Hopkins Bayview tells 3,182 patients an employee put their health data into a public AI tool
  2. China’s AI compute access expected to double in 2027, with US-designed chips about 65 per cent
  3. UK commits £100m to national AI Biosecurity Lab
  4. Moonshot AI releases Kimi K3.1 Pro with 1-million-token context window
  5. Residents in Newport and Cardiff oppose hyperscale datacentres near homes
  6. Study: Grok 4.7 picked Republicans in 92% of forced midterm answers, GPT-6.1 Sol picked Democrats in 97%
  7. Attackers exploiting the PHP flaws hand the break-in to AI agents, Unit 42 says
  8. Elvis Presley’s rightsholders sue Anthropic over training on 274 songs

Latest / Frontier models and safety

ReportFrontierScienceUSConfirmed

Anthropic: Claude writes 80% of its code; self-improvement could slip out of control

See how this story connects

Source: AnthropicAlso reported by Tom’s Hardware

The Anthropic Institute published “When AI builds itself”, by Marina Favaro and Jack Clark, on recursive self-improvement, the point at which a model designs and builds its successor without meaningful human input. It says that as of May 2026 Claude wrote more than 80% of the code merged into Anthropic’s production codebase; that Claude succeeded on 76% of the most open-ended coding tasks in May 2026; that on an internal test of speeding up training code, results rose from about 3x with Claude Opus 4 in May 2025 to about 52x with Mythos Preview in April 2026; and that Mythos Preview chose a better next research step than the human 64% of the time across 129 moments. It calls an 8x rise in code merged per engineer almost certainly an overstatement. It warns misalignment could compound “until we lose control of them”, says it would likely slow or pause only if other developers at or near the frontier did so verifiably, and calls for verification systems that would make a coordinated slowdown credible.

Why it matters

A frontier lab has published its own measurements of how far its models now build their successors, and named a coordinated pause as an option it would join.

When
4 Jun 2026
ID
WTA-N-2026-0285
Who
Anthropic; Jack Clark, Marina Favaro

Summary by wheresthe.ai, from Anthropic and the reporting credited above. Checked against the source on 13 Oct 2026.

Read the original at Anthropic

Source copy: download the page as saved on 18 Oct 2026 (PDF). Documents the source offers: document 1 (PDF), document 2 (PDF). Also held by the Internet Archive.

Next stop on this Storyline: Led toAnthropic says Claude now leads 26% of its AI research and development

Cite this story

WTA-N-2026-0285: “Anthropic: Claude writes 80% of its code; self-improvement could slip out of control.” wheresthe.ai, 4 Jun 2026. https://wheresthe.ai/d/anthropic-when-ai-builds-itself-2026-06/ (source: Anthropic).

Spotted an error? Email [email protected].