When AI Builds Itself
Anthropic, Thursday, June 4th, 2026
Anthropic demonstrates that AI systems are accelerating their own development toward recursive self-improvement.
Anthropic reports that AI systems are increasingly automating the AI development process itself, with Claude now authoring over 80% of merged code at the company and engineers shipping 8x more code per quarter compared to 2021-2025.
The company presents evidence that AI capabilities are advancing rapidly across benchmarks, with task completion times doubling every four months, and notes that while recursive self-improvement-where AI systems autonomously design and train their successors-is not yet achieved, current trends suggest it could arrive sooner than most institutions are prepared for.
This development presents both enormous potential benefits for science and healthcare, and significant risks regarding human control and oversight of increasingly autonomous AI systems. Anthropic's internal data shows AI has moved from suggesting code snippets to running and delegating work autonomously, though gaps remain in AI systems' ability to exercise judgment in choosing goals rather than executing specified tasks.