📊 Full opportunity report: When AI Builds Itself: Inside Anthropic’s Evidence on Recursive Self-Improvement on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Anthropic’s new report provides data indicating AI systems are now capable of automating significant parts of their own development. While human decision-making remains a bottleneck, the evidence suggests rapid progress toward autonomous AI improvement, though full self-advancement is not yet achieved.
Anthropic has released new evidence suggesting that AI systems are now capable of automating substantial parts of their own development, raising the possibility that recursive self-improvement could occur sooner than expected. The company’s internal data shows AI models like Claude are increasingly handling coding, experimentation, and even some research decisions, although human oversight remains essential. This development is significant because it indicates that AI could accelerate its own progress at a much faster rate, potentially transforming the landscape of AI research and deployment.
Anthropic’s recent report is based on internal data and public benchmarks that demonstrate AI models, especially Claude, are rapidly improving their ability to perform tasks traditionally done by humans in AI research and development. For example, by May 2026, over 80% of code merged into Anthropic’s projects was authored by Claude, up from a few percent in early 2025. Public benchmarks like METR show that AI’s capability to handle complex software tasks has doubled every four months, with models now able to perform tasks that previously required days, in hours or less. These trends suggest that AI is increasingly capable of automating both engineering and research activities, although the authors emphasize that the decision-making aspect—what problems to pursue—is still largely human-controlled.
The report clarifies that while AI’s technical capabilities are advancing rapidly, the critical bottleneck remains in the human judgment of which problems matter most. The authors highlight that AI models are already strong at executing specified experiments and coding tasks but are weaker at setting research goals and choosing which projects to prioritize. Despite this, the evidence indicates that if the human decision layer were to be automated, AI could potentially improve itself at a rate limited only by computational resources and safety considerations.
When AI builds itself
Anthropic is delegating a growing share of AI development to AI. Taken far enough, that points to a system that designs its own successor — recursive self-improvement. Not here yet, not inevitable. But the case isn’t speculation: it’s data on what AI is doing to AI development right now.
The curve that hasn’t bent
METR tracks the length of tasks AI can reliably complete on its own. That horizon is doubling roughly every four months — up from every seven. Anyone can check this in public data.
Task horizon — how long a job AI can handle solo
Each model handles dramatically longer tasks than the one a year before. The line keeps going up.
Coding with AI For Dummies (For Dummies: Learning Made Easy)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Two kinds of work, one persistent gap
Building a frontier model splits into engineering and research. Across both, the pattern is the same — and so is the one thing AI still can’t do well.
Code, infrastructure, training
Claude can take an underspecified problem and find a method. Humans supply the goal; they no longer need to supply the method.
Which experiments, what they mean
Claude can match or outperform skilled humans at executing a well-specified experiment. But choosing which experiment still needs a human.
The same ladder Anthropic employees climb with experience
CLAUDE AI UNLEASHED From First Prompts to Pro: The Complete Guide to Claude AI for Writing, Research, Coding, and Business (The Claude AI Mastery Series)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Watch the human share shrink, rung by rung
Walk up the four stages of AI development. At each, the human/AI split shifts — and the real internal numbers show exactly where AI has reached parity, gone superhuman, or still trails. Tap a rung.
The human role across the development loop
The doing now costs almost nothing in human time. What’s left is the deciding.
Agentic Development: The Complete Guide to AI-Assisted Coding with Claude, Cursor, and Beyond
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Agents ran an open research project end to end
April 2026: the first demonstration of Claude running an open-ended research project from hypotheses to findings — on a real AI-safety problem.
Can a weaker model reliably supervise a stronger one?
Agents were left to solve it: proposing hypotheses, testing them, sharing findings across parallel agents, iterating. Measured against the gap between a “floor” (weak supervisor alone) and “ceiling” (strong model trained on correct answers).
(humans: ~23% in a week)
· ~$18,000 compute
the agents themselves
AI-Assisted Programming: Better Planning, Coding, Testing, and Deployment
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Picking a better next step than the human
Real research sessions where a human took a wrong turn. Models saw only the work before the detour and proposed a next step; a judge that knew the outcome scored them. The day-to-day of research is this chain of next-step calls.
“Can the model pick a better next step than the human?”
Share of moments where the model’s next step was judged better. The amber line is the practical ceiling (an ideal answer that could see the whole session).
It depends on whether the trend continues — and what we do
The piece refuses a single prediction. It lays out three scenarios, and is clear about which it finds most likely.
The exponentials turn out to be S-curves
Maybe taste can’t be scaled into existence; maybe the constraint is the supply chain — chips, grid, interconnect — not intelligence. Even so, the world still changes: Glasswing’s Mythos found 10,000+ critical vulnerabilities in weeks, and a 100-person firm does the work of 1,000.
included for completeness · they doubt itDevelopment automates; humans still steer
100-person companies doing the work of tens of thousands — revolutionary, but turnable to harm (population-scale surveillance, tailored manipulation). Bound by Amdahl’s law: speeding one part shifts the bottleneck — which is exactly why human code review became Anthropic’s new chokepoint.
★ they think we’re likely heading hereAI designs and refines its own successors
Progress paced only by compute. Humans move to oversight of an expanding “virtual lab.” The future they understand least — especially whether alignment holds, or whether rare misalignments compound as models build successors, until control slips.
the one they’re most uncertain aboutBuild the option to slow down — verifiably
The piece ends on policy, not product. A unilateral pause just changes who leads; what’s missing is the ability to verify others have actually slowed.
Why a credible pause is hard — and worth building toward
A slowdown that only lets the least cautious catch up leaves everyone less safe. So the goal is the option: systems that let frontier labs verify others have genuinely stopped. Anthropic says if such systems existed and peers paused verifiably, it expects it would too.
Detection beats verification — and even that’s tough
Training runs are easier to conceal than missile silos, inputs are general-purpose, and whoever continues while others pause inherits the lead.
We’ve done it before — slowly
Regimes like the INF Treaty built verification and trust over decades. The authors’ blunt line: “We don’t have that long.”
Reading it in proportion
- This is one lab’s account of its own internal data — much previously unreported, not independently audited.
- The soft spots are stated in the original: lines-of-code overstates productivity; the self-reported 4× is probably high; the headline research result didn’t transfer to production scale; the next-step test used cherry-picked moments.
- “More autonomous” is not “fully autonomous” — every standout result still had a human framing the problem and defining success.
- That the authors surface these caveats themselves — against their own incentive — is part of what makes the document serious.
Implications of AI Automating Its Own Development
This evidence suggests that AI systems are moving toward a state where they could independently design and improve themselves, a concept known as recursive self-improvement. If achieved, this could lead to rapid advancements in AI capabilities, potentially surpassing human control or oversight in certain areas. Such progress raises important questions about safety, governance, and the future of AI development, as the pace of innovation could accelerate beyond current regulatory or ethical frameworks.
Recent Advances in AI Development and Benchmarks
Over the past few years, AI models have shown consistent improvements across various benchmarks. Public data from metrics like METR and SWE-bench indicate that AI’s ability to perform complex software tasks, fix bugs, and reproduce research results has doubled roughly every four months. Internally, Anthropic’s data reveals that Claude’s coding output has increased dramatically, with over 80% of code in recent projects authored by the model. These trends mirror broader industry patterns of rapid AI capability growth, driven by larger models, better training techniques, and increased compute resources.
However, most of these benchmarks measure task performance, not the internal pace of AI-driven research. The internal data from Anthropic provides rare insight into how AI might be automating the research process itself, including coding, experiment execution, and even some aspects of problem selection. This internal evidence forms the core of the report’s argument that the next frontier is AI systems autonomously improving their own capabilities.
“The data from Anthropic suggests we are witnessing the early stages of AI systems automating their own research and development, which could accelerate progress dramatically.”
— Thorsten Meyer, AI researcher
Uncertainties Surrounding AI Self-Improvement Pace
It remains unclear whether AI will soon be able to fully automate the research decision layer, which involves setting goals, prioritizing problems, and evaluating results. The current evidence shows rapid technical progress but does not confirm that AI can autonomously design its own successors or improve itself without human input. Additionally, safety, ethical, and governance considerations could slow or restrict such developments, and it is uncertain when or if these barriers will be overcome.
Next Steps in Monitoring AI Self-Development
Researchers and industry observers will continue to analyze internal data from AI labs like Anthropic and monitor public benchmarks for signs of increased autonomy. Regulatory and safety frameworks are also likely to evolve, aiming to address the potential risks of autonomous AI self-improvement. Further transparency from labs about internal metrics and progress will be crucial to understanding how close AI systems are to achieving recursive self-improvement and what safeguards might be necessary.
Key Questions
What is recursive self-improvement in AI?
Recursive self-improvement refers to an AI system’s ability to autonomously improve its own capabilities, potentially leading to rapid, exponential growth in intelligence and performance without human intervention.
How does Anthropic’s internal data support claims of AI self-improvement?
Anthropic’s internal metrics show that AI models like Claude are increasingly handling tasks such as coding and experimentation, with significant growth in their autonomy and output, suggesting progress toward self-improvement capabilities.
What are the risks associated with AI self-improvement?
Potential risks include loss of human control, unpredictable behavior, and safety concerns if AI systems begin to self-modify beyond current oversight. These issues are still under active discussion among researchers and regulators.
Is AI currently capable of fully automating its own research and development?
No, current evidence indicates AI can automate many technical tasks but still relies on human judgment for goal-setting and strategic decisions. Full automation of self-improvement remains a future possibility, not an immediate reality.
Source: ThorstenMeyerAI.com
