📊 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 reports measurable acceleration in AI’s ability to perform research tasks, with data indicating that AI systems are now automating parts of AI development. While promising, the crucial step of autonomous goal-setting remains unachieved. This could accelerate AI progress significantly if fully realized.
Anthropic has published new evidence suggesting that AI systems are now capable of automating significant portions of AI research and development, potentially enabling recursive self-improvement if certain bottlenecks are overcome. This development is based on internal data and public benchmarks, marking a rare data-driven insight into current AI progress toward self-automating research capabilities.
The report from The Anthropic Institute indicates that AI models, especially Claude, are increasingly performing tasks traditionally done by human researchers, such as writing code, running experiments, and analyzing results. Notably, Anthropic engineers now ship eight times more code quarterly than they did between 2021 and 2025, with models like Claude responsible for over 80% of code merges as of May 2026.
Public benchmarks such as METR, SWE-bench, and CORE-Bench show rapid improvements in AI capabilities, with models now handling tasks ranging from bug fixing to reproducing research results. For example, Claude Mythos Preview can work on tasks lasting at least 16 hours, a significant increase from earlier capabilities.
Inside labs, data suggests that AI systems are climbing the ladder of research tasks—from executing specific tasks to designing experiments—though they still lag in higher-level decision-making, such as choosing research goals. The authors emphasize that while progress is clear at lower levels, autonomous goal-setting remains unachieved, representing the key gap toward recursive self-improvement.
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 research code development 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 experiment 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 programming and coding assistants
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
Evals for AI Engineers: Systematically Measuring and Improving AI Applications
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 Impact of AI Self-Research Capabilities
This evidence indicates that AI systems are already automating parts of the research process, which could accelerate AI development dramatically if the ability to set and pursue research goals autonomously is achieved. Such a shift would mean AI could improve itself at a faster pace than human-led methods, raising questions about the future control and safety of AI systems.
While the current progress is promising, experts caution that the critical bottleneck—autonomous goal-setting—remains unresolved. If overcome, it could lead to a rapid, self-perpetuating cycle of AI improvement, with profound implications for technology, regulation, and safety.
Current State of AI Self-Improvement Research
Anthropic’s report is among the first to use internal data and public benchmarks to measure AI’s progress in automating research tasks. Historically, AI development has relied heavily on human researchers designing experiments and setting goals. Recent advancements, however, suggest models like Claude are now capable of executing complex research tasks with minimal human input.
Benchmarks such as METR, SWE-bench, and CORE-Bench demonstrate rapid capability improvements over the past two years, with models handling increasingly complex and longer-duration tasks. Despite this progress, the ability of AI to autonomously define research objectives remains a significant challenge, keeping the potential for recursive self-improvement theoretical rather than practical at this stage.
“Our data shows that AI models are now performing a substantial portion of the research work that was previously done exclusively by humans, which is a significant step toward autonomous AI research.”
— Thorsten Meyer, lead author of the report
Unresolved Challenges in AI Autonomous Goal-Setting
It is not yet clear whether AI systems can autonomously define meaningful research objectives at scale. The current evidence shows progress in executing tasks but does not confirm that models can independently pursue improvements without human guidance. The timeline and technical feasibility of achieving this remain uncertain, and some experts warn it may be more difficult than current capabilities suggest.
Next Steps in Monitoring AI Self-Research Development
Researchers and industry observers will closely track further internal data from labs like Anthropic, as well as public benchmark progress, to assess whether AI can reach autonomous goal-setting. Future work will likely focus on developing and testing models specifically designed to pursue self-improvement goals, with safety and control considerations remaining central. Regulatory and safety frameworks may also evolve in response to these developments.
Key Questions
What does recursive self-improvement mean in AI?
It refers to AI systems being able to improve their own design and capabilities without human intervention, potentially leading to rapid, exponential progress.
Is AI currently capable of fully automating its own research?
No, current AI systems are automating parts of research tasks, but the ability to autonomously define research goals remains unachieved.
Why is autonomous goal-setting important for self-improvement?
Because without the ability to set and pursue their own research objectives, AI systems cannot independently improve themselves beyond executing predefined tasks.
What are the risks if AI achieves recursive self-improvement?
Potential risks include loss of human control, unpredictable behavior, and rapid technological changes that could outpace safety measures and regulation.
Source: ThorstenMeyerAI.com
