Anthropic reveals Claude writes 80%+ of its own code – and warns recursive self-improvement could arrive sooner than the world is ready
Category: Ai Safety
What it is
On June 4, the Anthropic Institute published “When AI Builds Itself,” disclosing that
Claude now authors over 80% of code merged into Anthropic’s production systems – up
from low single digits before Claude Code launched in February 2025. Engineers now
merge 8× more code daily than in 2024, and the Mythos Preview model hit a 52× code
speedup benchmark vs. 4× for skilled humans. Co-founder Jack Clark told the BBC that
100% AI-written code is possible within two years and called for a coordinated global
“brake pedal” before then, noting that a one-sided pause would simply let rivals race
ahead.
Why it Matters for Enterprises
This is the clearest evidence yet that AI is compressing software R&D timelines at an
unprecedented rate. For enterprise technology leaders, the productivity gains are already
real - but so is the need to build governance frameworks for AI-authored code before the
first production incident, not after.
