📊 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 models are now capable of automating significant parts of AI research and development. The key question is whether this progress could lead to recursive self-improvement, where AI enhances itself without human intervention. The evidence is based on internal metrics and public benchmarks, but some gaps remain.
Anthropic’s new report indicates that AI systems are capable of automating substantial parts of AI research and development, with data showing notable progress in capabilities. The report discusses the potential for AI to automate the final human decision-making step—research taste—and the implications this could have for recursive self-improvement.
The report from The Anthropic Institute highlights that AI models, particularly Claude, are now performing tasks previously requiring human expertise at an increasing rate. Metrics from public benchmarks reveal that AI’s ability to handle complex coding, debugging, and research tasks has improved significantly over the past year, indicating rapid progress.
Internal data from Anthropic shows that over 80% of the code merged into its projects as of May 2026 was authored by Claude, a notable increase from previous years. Additionally, models like Claude Mythos Preview can now work on tasks lasting at least 16 hours, approaching the range of complex, multi-day research projects.
The report distinguishes between engineering—writing code and infrastructure—and research—deciding what experiments to run. While AI has made significant strides in engineering tasks, gaps remain in AI’s ability to autonomously select research goals and prioritize problems, which the authors identify as key bottlenecks for true 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.
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.
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 debugging tools
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-Driven Self-Improvement
This development suggests that AI could increasingly automate not only coding and experimentation but also aspects of research decision-making. If AI can autonomously design its successors, it may influence the pace of innovation. The findings highlight the importance of ongoing discussions about oversight, safety, and regulation to manage these technological advancements.
Current State of AI Capabilities and Benchmarks
Over the past few years, AI models have shown consistent improvement in public benchmarks like METR, SWE-bench, and CORE-Bench, which evaluate capabilities such as code generation, bug fixing, and reproducing research results. These metrics illustrate a steady upward trajectory, with models now handling tasks that previously required days of human effort in hours or less. Internal data from Anthropic further confirms that AI is increasingly performing core research and development activities, indicating a trend toward greater automation in AI labs.
Progress has been steady, but recent data suggests the potential for acceleration, especially in the ability of models to perform complex tasks autonomously. This trend supports the hypothesis that AI might soon reach a point where it can improve itself without human input, provided the bottleneck of research taste can be addressed.
“The evidence from Anthropic indicates that AI is already taking on roles in research and development that were previously thought to require human judgment, raising important considerations for future development.”
— Thorsten Meyer, AI researcher
Unresolved Questions About AI Self-Improvement Potential
It remains uncertain whether current trends will continue at the same pace or level off. The primary unknown is whether AI can fully automate the decision-making process—research taste—that guides problem selection, which is currently managed by humans. Additionally, discussions around the safety, ethical, and regulatory implications of potential recursive self-improvement are ongoing, with no consensus on timelines or risks.
Next Steps in Monitoring AI Development and Safety
Further research, both internally and externally, will be necessary to assess whether AI can autonomously design and improve its own systems at scale. Industry and regulatory bodies are expected to increase scrutiny of these developments, potentially leading to new standards for AI safety and oversight. Researchers will also focus on addressing the gap in autonomous goal-setting to better understand the feasibility of recursive self-improvement.
Key Questions
What is recursive self-improvement in AI?
Recursive self-improvement refers to AI systems that can autonomously enhance their own design and capabilities, potentially leading to rapid increases in intelligence without human intervention.
How does Anthropic measure AI’s progress in research tasks?
Anthropic uses benchmarks such as METR, SWE-bench, and CORE-Bench, which evaluate AI models’ abilities to perform coding, debugging, and research reproduction tasks, tracking improvements over time.
What are the risks of AI achieving recursive self-improvement?
Potential risks include loss of human oversight, unpredictable behavior, and safety challenges, which could complicate control and regulation of increasingly autonomous AI systems.
Is AI currently capable of designing its own successors?
There is currently no conclusive evidence that AI can fully autonomously design and improve its own systems. While progress is ongoing, key challenges in autonomous goal-setting remain.
Source: ThorstenMeyerAI.com
