📊 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 that AI models are now capable of automating significant parts of AI development, with potential for self-improvement. While current evidence shows progress, the leap to fully autonomous self-improvement remains unconfirmed and uncertain.
Anthropic’s latest report reveals that AI systems are increasingly capable of automating core research and engineering tasks, raising the possibility that AI could soon improve itself without human intervention. While the report stops short of claiming that recursive self-improvement has been achieved, it provides concrete data suggesting the development process is accelerating faster than most expected, which could have profound implications for AI safety and governance.
The report from Anthropic’s Institute presents internal and public benchmarks showing that AI models like Claude now perform a growing share of coding and experimental tasks. For example, more than 80% of code merged into Anthropic’s codebase by May 2026 was authored by Claude, up from single digits in early 2025. Public benchmarks such as METR indicate that AI’s ability to handle complex tasks has doubled roughly every four months, suggesting rapid progress in AI capabilities.
Internally, the report distinguishes between engineering—writing code and infrastructure—and research—deciding experiments and interpreting results. It notes that while AI can now match or outperform skilled humans at executing specific experiments, significant gaps remain in AI’s ability to autonomously choose research goals or determine which problems are worth pursuing. The authors emphasize that current evidence shows progress up the ladder of research autonomy, but full autonomous self-improvement is not yet realized.
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.
AI coding assistant tools
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
AI research automation software
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.
AI development environment
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 programming IDE
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.
Potential for AI to Self-Improve Accelerates Development
This development matters because it suggests that AI systems could soon reach a point where they can improve their own design and research processes with minimal human input. Such a shift could drastically speed up AI progress, potentially leading to rapid, recursive improvements. This raises important questions about control, safety, and the future pace of AI innovation, which are central to ongoing debates in AI policy and ethics.
Evidence of Rapid Progress in AI Research and Engineering
Anthropic’s report builds on recent public benchmarks and internal data showing AI models’ capabilities are growing at an unprecedented rate. The METR benchmark, which measures AI’s ability to perform tasks like coding and debugging, has shown a doubling of capabilities every four months. Internal data indicates that AI is now responsible for a majority of new code contributions and can perform complex research tasks, such as reproducing results or fixing bugs, at levels comparable to skilled humans.
While these trends are clear, experts caution that the evidence is still limited to specific tasks and capabilities, and do not yet confirm that AI can fully autonomously design or improve its own systems without human oversight.
“The data from Anthropic indicates that AI is already automating significant parts of the research and development process, which could accelerate toward full self-improvement if certain bottlenecks are overcome.”
— Thorsten Meyer, AI researcher
Unconfirmed Potential for Fully Autonomous Self-Improvement
It remains unclear whether AI will soon be able to autonomously identify, design, and implement improvements to itself without human guidance. The evidence shows rapid progress in task execution but does not confirm that the critical step—self-directed goal setting—is imminent or inevitable. Experts warn that significant technical and safety challenges must still be addressed before such capabilities could be realized.
Monitoring Progress and Addressing Safety Challenges
Researchers and policymakers will closely watch ongoing developments in AI capabilities and benchmark progress. Focus will also be on establishing safety protocols and governance frameworks to manage the risks associated with increasingly autonomous AI systems. Further internal data from labs like Anthropic will be crucial to understanding if and when recursive self-improvement becomes feasible.
Key Questions
What is recursive self-improvement in AI?
Recursive self-improvement refers to an AI system’s ability to autonomously improve its own design and capabilities, potentially leading to rapid, exponential progress without human intervention.
Has AI already achieved self-improvement capabilities?
Currently, there is no confirmed evidence that AI can fully autonomously improve itself. Evidence suggests progress in automating research tasks, but the key step—self-directed goal setting—is still human-controlled.
Why does this development matter for AI safety?
If AI systems can self-improve rapidly, it could lead to unpredictable and uncontrollable growth in capabilities. Ensuring safety and alignment becomes critical as the pace of development accelerates.
What are the main limitations of current evidence?
Most data pertains to specific tasks and benchmarks, not the full spectrum of AI research and development. The leap to fully autonomous self-improvement involves challenges not yet demonstrated or understood.
What should we expect next from research labs?
Ongoing benchmarking, internal data collection, and safety research will continue. Stakeholders will monitor whether AI systems can autonomously progress beyond current capabilities and address associated risks.
Source: ThorstenMeyerAI.com
