📊 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 presents data indicating AI systems are already automating parts of AI research and development. If human oversight of goal-setting and decision-making diminishes, AI could begin self-improving at unprecedented speeds.
Anthropic’s new report provides the first concrete evidence that AI models are increasingly automating parts of AI research and development, raising the possibility of recursive self-improvement if human oversight of goal-setting and decision-making diminishes.
The report, published by The Anthropic Institute, details internal data showing that AI models like Claude are now capable of writing significant portions of code, running experiments, and producing results with minimal human input. For example, over 80% of code merged into Anthropic’s codebase as of May 2026 was authored by Claude, up from single digits in early 2025.
Public benchmarks such as METR, SWE-bench, and CORE-Bench demonstrate rapid progress in AI’s ability to perform tasks traditionally requiring human expertise, with capabilities doubling every four months on some measures. These trends suggest that AI could soon handle tasks that currently take humans days or weeks, within months or a few years.
Despite these advances, the report emphasizes that the most significant gap remains in AI’s ability to choose research goals and prioritize problems—an area where human judgment still dominates. The authors argue that if this bottleneck also becomes automatable, a feedback loop of self-improvement could emerge.
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.
Agentic Coding with Claude Code: The everyday developer's guide to agentic coding with Claude Code
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
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.
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
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 Accelerated AI Self-Improvement
This development matters because it indicates that AI systems are not only improving in capability but are also increasingly capable of automating the process of AI research itself. If the bottleneck of human decision-making in research is eliminated, AI could enter a loop of rapid self-enhancement, dramatically accelerating technological progress and raising questions about control and safety.
Such a shift could impact AI development timelines, regulatory considerations, and the future of AI safety protocols, making it a key area for ongoing monitoring and discussion among researchers and policymakers.
Data-Driven Evidence of AI Development Acceleration
The report builds on public benchmarks showing AI models’ capabilities doubling every four months, with models like Claude reaching tasks previously requiring hours or days in a matter of months. Internal data from Anthropic further supports this trend, revealing that the proportion of code authored by AI has surged dramatically in less than two years.
Historically, progress in AI has been measured by performance on benchmarks, but this report emphasizes internal metrics that track how AI is contributing to its own development—an area less visible to outsiders but critical for understanding potential self-improvement.
While the evidence is compelling, experts caution that the key unknown remains whether AI can fully automate the decision-making process—particularly goal-setting—that guides research and development.
“The internal data from Anthropic shows a clear trend: AI is increasingly capable of automating significant portions of its own development, but the critical bottleneck—deciding what to improve—is still human-controlled.”
— Thorsten Meyer, AI researcher
Unconfirmed Aspects of Autonomous Self-Improvement
It is not yet clear whether AI systems will be able to fully automate the goal-setting and strategic decision-making processes necessary for recursive self-improvement. The evidence shows progress in technical capabilities but does not confirm that AI can autonomously design and prioritize its own research agenda without human input. The timeline and feasibility of this transition remain uncertain.
Monitoring Progress Toward Autonomous AI Self-Enhancement
Researchers and institutions will likely focus on developing benchmarks and internal metrics to better measure AI’s ability to automate research decision-making. Further internal disclosures from labs like Anthropic could clarify whether the bottleneck of human oversight diminishes naturally or requires deliberate engineering. Policy discussions around AI safety and control will intensify as capabilities continue to evolve.
Key Questions
What is recursive self-improvement in AI?
Recursive self-improvement refers to an AI system’s ability to autonomously enhance its own capabilities, potentially leading to rapid and exponential growth in intelligence and performance.
How does Anthropic’s internal data support the idea of AI self-improvement?
Anthropic’s internal metrics show AI models increasingly automating tasks like coding and experimentation, suggesting that parts of AI development are already being handled by AI systems themselves.
What are the main obstacles to AI achieving full self-improvement?
The primary obstacle is the human-controlled decision-making process—specifically, the ability of AI to autonomously set research goals and prioritize problems, which remains largely in human hands.
Why is this development significant for AI safety?
If AI can self-improve rapidly without human oversight, it raises questions about control, predictability, and safety, making it a critical area for ongoing research and regulation.
When might AI reach the point of autonomous self-improvement?
It is currently uncertain; while the evidence suggests rapid progress, the timeline for AI to fully automate research decision-making remains unknown and is subject to ongoing developments.
Source: ThorstenMeyerAI.com
