When AI builds itself
Our progress toward recursive self-improvement, and its implications.
For most of AI’s history, humans drove every step in its development cycle. But at Anthropic, we are delegating a growing share of AI development to AI systems themselves, which is speeding up our work.
Taken far enough, and given enough compute, that trend points to an AI system capable of fully autonomously designing and developing its own successor. This is called recursive self-improvement. We are not there yet, and recursive self-improvement is not inevitable. But it could come sooner than most institutions are prepared for.
Using public benchmarks and previously unreported data from within Anthropic, The Anthropic Institute is showing that AI is already accelerating the development of AI systems. To take just one example: today, Anthropic engineers on average ship 8x as much code per quarter as they did from 2021-2025.
The technical trends discussed in this piece suggest that AI systems are going to become much more capable in coming years. These trends have huge implications. AI that can build itself would be a major development in the history of technology—one that could bring enormous good for the world in science, healthcare, and beyond. But full recursive self-improvement also might increase the risks of humans losing control over AI systems. If systems are capable of fully building their own successors, the ways we secure them, monitor them, and shape their behavior all grow much more important.
Timeline of AI-assisted development at Anthropic
- 2021–2023: Building the first Claude — work looked like any other tech company: people writing code and docs on laptops.
- 2023–2025: Chatbots — people used early chatbots to help with parts of the process, like generating short code snippets.
- 2025–2026: Coding agents — agents could write and edit code on their own, sometimes entire files.
- Today: Autonomous agents — agents can run code themselves and delegate hours of work to other agents.
- Future?: Closing the loop — agents could become capable enough to build and train models themselves.
Evidence from the outside world
The rate at which AI models improve is accelerating. The length of tasks that they can reliably complete on their own has been doubling roughly every four months, up from an earlier trend of doubling every seven months. In March 2024, Claude Opus 3 could complete software tasks that take humans about four minutes. A year later, Claude Sonnet 3.7 managed tasks that took about an hour and a half. A year after that, Claude Opus 4.6 managed 12-hour tasks. If this trend holds, tasks that take a skilled person days could come into range this year. In 2027, AI systems could be capable of tasks that take a person weeks.
Benchmarks like SWE-bench and CORE-Bench have saturated in roughly two years as models went from low single-digit scores to near-100% performance. METR found Claude Mythos Preview could work for “at least” 16 hours.
Evidence from within Anthropic
Building a frontier model takes engineering (code, infrastructure, training oversight) and research (experiments, interpretation, next ideas).
Code authorship: As of May 2026, more than 80% of the code merged into Anthropic’s codebase was authored by Claude. Before Claude Code launched in research preview in February 2025, this number was in the low single digits. In Q2 2026, the typical engineer was merging 8× as much code per day as in 2024.
Code quality: On the most open-ended tasks, Claude’s success rate reached 76% in May 2026, up 50 percentage points in six months. Many staff believe Claude-written code was worse than human-written code in late 2025, is roughly at parity today, and will be better within the year.
Automated review: An automated Claude reviewer now reads proposed codebase changes before merge and would have caught roughly a third of bugs behind past claude.ai incidents.
Research automation: In a miniature training-speedup test, Claude went from ~3x speedup (May 2025, Opus 4) to ~52x (April 2026, Mythos Preview). Claude agents also ran an open-ended AI safety research project recovering 97% of a performance gap that took two human researchers a week to achieve 23% of, over 800 cumulative hours and ~$18,000 compute.
Research judgment: On 129 challenging next-step decisions in real research sessions, Opus 4.5 beat the human choice 51% of the time in November 2025; Mythos Preview reached 64% in April 2026.
Possible futures
- Trend stalls but capabilities diffuse widely — S-curves, supply-chain constraints, or exogenous shocks slow progress; today’s models still transform economies.
- Compounding efficiency gains — AI development substantially automated; humans set direction; 100-person companies do work of 10,000–100,000-person organizations; new bottlenecks emerge (e.g., human code review).
- Full recursive self-improvement — AI systems design successors; pace set by compute; humans shift to oversight of a “virtual lab”; alignment outcomes highly uncertain.
Policy recommendations
Anthropic argues the world should have the option to slow or temporarily pause frontier AI development so societal structures and alignment research can keep pace. A meaningful pause requires verifiable multi-lab, multi-country coordination — harder than nuclear arms control because training runs are easier to conceal.
In the coming months, Anthropic will organize conversations with policymakers, researchers, civil society, and other AI companies on recursive self-improvement and coordination mechanisms, and publish outcomes.
“We believe it would be good for the world to have the option to slow or temporarily pause frontier AI development to enable societal structures and alignment research to keep up with the advance of the technology.”