📊 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 internal data shows AI systems are accelerating their own development, with models increasingly automating coding and research tasks. While current evidence suggests progress, full autonomous self-improvement remains unconfirmed and uncertain.
Anthropic’s new report presents evidence that AI systems are increasingly capable of automating their own development, with models now performing significant portions of coding and research tasks without human input. This suggests that the possibility of recursive self-improvement — AI improving itself at a rapid pace — may be closer than previously thought, though it is not yet realized or inevitable.
The report, published by The Anthropic Institute, bases its findings on public benchmarks and internal data, showing that AI models like Claude have dramatically increased their ability to generate code and perform research tasks. For example, over the past 15 months, the proportion of code written by Claude in Anthropic’s projects has risen from single digits to over 80%, indicating rapid automation of engineering work.
Public data shows that AI capabilities are doubling roughly every four months in terms of task complexity, with models now handling tasks that previously required days of human effort. Benchmarks such as METR and CORE-Bench demonstrate that models are closing in on fully autonomous research and development, with some models managing tasks spanning hours to days. However, internal data indicates that models still lack the ability to independently decide research goals or prioritize problems, which remains a significant gap before true recursive self-improvement can occur.
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 automation 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
machine learning code generator
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 Accelerating AI Self-Development
This evidence suggests that AI systems are rapidly advancing in automating parts of their own development process, which could shorten the cycle time for innovation and research. If models eventually reach a point where they can autonomously set goals, design experiments, and improve their own architecture, it could lead to a feedback loop of rapid self-improvement. Such a development would have profound implications for AI safety, regulation, and the future of technological progress, making it a critical area for ongoing monitoring and research.
Current State of AI Self-Improvement Capabilities
While AI models like Claude have shown remarkable progress in automating coding and research tasks, the idea of recursive self-improvement remains theoretical. Historically, AI progress has been measured through benchmarks and internal metrics, but the internal data from labs like Anthropic now indicates that models are approaching capabilities that could enable more autonomous development. However, the critical gap is in the models’ ability to independently decide which research problems to pursue, an area still dominated by human judgment.
“The data from Anthropic suggests that AI is already automating significant parts of its own development, but the leap to full self-directed improvement is still unconfirmed.”
— Thorsten Meyer, AI researcher
Uncertainties Surrounding Autonomous Self-Improvement
It remains unclear whether current trends will continue at the same pace, and whether models will ever reach the capability to autonomously set research goals and improve themselves without human oversight. The authors acknowledge that this is a conditional possibility, dependent on future developments in AI architecture and training, and that there are significant technical and safety challenges ahead.
Next Steps in Monitoring AI Self-Development
Researchers and industry observers will likely focus on tracking internal model capabilities, especially in goal-setting and decision-making. Further experiments and transparency from labs like Anthropic will be critical to understanding whether recursive self-improvement is feasible and imminent. Policy discussions around AI safety and oversight are expected to intensify as models approach higher levels of autonomy.
Key Questions
What is recursive self-improvement in AI?
Recursive self-improvement refers to AI systems that can autonomously enhance their own capabilities, potentially leading to rapid, exponential progress without human intervention.
How does Anthropic measure AI’s progress in self-improvement?
Anthropic uses public benchmarks like METR and CORE-Bench, along with internal data on coding output and task performance, to track how AI models automate research and development tasks.
Is AI currently capable of fully self-improving without human input?
No. While models are automating many tasks involved in AI development, they still lack the ability to independently set research goals or decide which problems to pursue, which is a key step toward true self-improvement.
What are the risks if AI begins to self-improve rapidly?
Rapid self-improvement could lead to unpredictable behavior, safety concerns, and challenges in regulation, making it crucial to monitor progress and develop safety protocols.
When might autonomous self-improvement become a reality?
It is uncertain; current data suggests it could happen sooner than most expect if technical and safety hurdles are overcome, but no definitive timeline exists.
Source: ThorstenMeyerAI.com
