🔍 Read the full analysis: Has Recursive Self-Improvement Started? What The Evidence Shows, And What A Ban Would Cover on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Evidence indicates that AI tools are helping human researchers accelerate parts of AI development, but no documented system has independently designed and built its successor. A proposed US bill would pause some AI development, including recursive self-improvement, but remains in committee and has no effect on businesses or users now.
AI is assisting the people who develop AI, and one independent evaluator says Anthropic’s progress on certain tasks has sped up, but available evidence does not show a system autonomously building its own successor. The debate has gained attention this week after podcaster Dwarkesh Patel published an essay on November 13; a separate proposal in Congress would pause some AI development, but both relevant bills remain in committee.
The distinction at the center of the debate is between AI helping human researchers and AI independently carrying out the research and development needed to build a more capable successor. Anthropic’s August 2026 figures put Claude at the “leads” level for 26% of the company’s AI research and development work. The company says more than 90% of the work it measured involved AI at least at a collaboration level, but no measured subset was fully autonomous.
Independent evaluator METR assessed Anthropic using internal data from August 12 to October 29, 2026. Its November 5 report said the company was likely improving models on well-specified tasks at around twice the rate seen between March 2024 and October 2025. METR emphasized that estimate had high uncertainty, giving a possible range from about 1.3 times to three times the earlier rate. It attributed most of the acceleration to AI assistance, not fully autonomous contributions.
METR also said AI was not yet making highly significant, fully autonomous contributions to the most important parts of model training. It warned that Anthropic had redacted much of the evidence the evaluator considered most informative. Separately, The Decoder reported on figures in an OpenAI blog post, including an “automated research intern” milestone and a goal of a fully automated AI researcher by March 2028. Those figures came from OpenAI’s own measurements, and the account said no independent review was cited.
Has it started? · explainer
AI already does much of the work of building AI. Not alone.
How far AI research and development is automated at one lab, on a scale from AL0 (no AI) to AL5 (no human in the loop), and where the proposed ban would draw its line.
Anthropic’s AI R&D work, August 2026
Automation levels from Epoch AI’s scale, as measured by Anthropic
Anthropic: “We are not there yet.”
- AI collaborates on most of the workAbove 90% of the work is at or above “AI collaborates”, where AI does large chunks under close human direction.
- AI leads about a quarterIn 26%, Claude completes most of a task from a high-level prompt while a human supervises.
- No closed loopNo measured part runs fully autonomously. That is the step “recursive self-improvement” usually means.
- Where would a ban start?The bill names “greatly accelerate” as a warning sign, so its reach depends on a measurement nobody yet makes in a standard way.
Sources: Anthropic, “Measurements for understanding the pace of AI development inside frontier labs” and “When AI builds itself” (figures as of August 2026, Anthropic’s own measurements); Ban Artificial Superintelligence Act of 2026, S.5493 and H.R.10538, sec. 3(6)(A). Bars show shares of Anthropic’s AI R&D work.
Graphic: Powered by Thorsten Meyer AI
Why Human Oversight Still Matters
The difference between AI-assisted research and autonomous self-improvement matters for both safety and policy. If systems can reliably conduct more of the work involved in developing future systems, researchers could make progress faster, potentially leaving less time for testing, oversight and governance. But the evidence described so far does not establish that AI has taken control of that development loop.
The distinction also affects the scope of proposed regulation. A rule covering any system that can accelerate AI research could reach tools used under human supervision, not only a hypothetical system that creates its own successor without people. Readers should not treat evidence of faster, AI-assisted work as proof that the strongest version of recursive self-improvement has arrived.
Products Worth Considering

Samsung 27-Inch Odyssey OLED G5 QHD Gaming Monitor (G50SF)

Apple Watch Series 12 GPS 46mm Smartwatch

Anker SOLIX C1000 Gen 2 Portable Power Station with 200W Solar Panel
As an affiliate, we earn on qualifying purchases.
The Bills Behind the Debate
The Ban Artificial Superintelligence Act of 2026 was introduced by Sen. Bernie Sanders as S.5493 on September 23 and by Rep. Greg Casar as H.R.10538 on September 24. The bills would pause training, modification and fine-tuning of large AI systems until a new Department of Artificial Intelligence establishes rules. Their text includes recursive self-improvement and would also prohibit systems with the capacity to automate or greatly accelerate AI research and development.
As of the source’s November 16 account, both proposals had been referred to committee and had not advanced. The House bill had 12 Democratic cosponsors. A separate proposal from Rep. Ro Khanna, called the Human Control Over AI Act, was described in a September CNBC report, but the source said it could find no bill text. The recent political attention follows Democrats’ wins in both chambers of Congress on November 3, according to the Associated Press; that outcome does not itself change the bills’ status.
““We are not there yet, and recursive self-improvement is not inevitable.””
— Anthropic, in its essay “When AI Builds Itself”
Limits of the Available Evidence
The evidence remains incomplete. METR said its estimate carried high uncertainty and that Anthropic redacted much of the material most useful for evaluating its conclusions. The reported OpenAI figures also come from the company itself, with no independent review cited in The Decoder’s account. Neither dataset, as described in the source material, establishes how well the results generalize beyond the companies’ own research environments.
It is also not clear how broadly the proposed ban’s terms would apply in practice, since the bills have not moved through committee and no implementing rules exist. The source reports no current legal change for companies or users. Whether Congress will revisit the proposals, and whether any revised version would distinguish between human-supervised AI tools and autonomous systems, remains unsettled.
Committee Action and New Evaluations
The immediate next steps are congressional consideration of the two introduced bills and further evidence about AI’s role in research and development. The proposals would need to advance beyond committee before becoming law; no such change has happened. OpenAI’s stated target of a fully automated AI researcher by March 2028 is a company goal, not a confirmed achievement or independent forecast.
Further evaluations could clarify how much AI assistance contributes to faster model development and whether systems can take on more consequential work without human intervention. For now, the most supportable conclusion is narrower: the AI-assisted loop is operating with people still involved, while fully autonomous successor-building has not been documented in the material reviewed.
Key Questions
Has recursive self-improvement started?
AI is helping people develop AI, which can be described as a human-supervised part of the loop. The source material does not document an AI system independently designing and building its own successor.
What did METR find about Anthropic?
METR estimated that Anthropic’s progress on well-specified tasks was around twice its earlier rate, while stressing high uncertainty and a possible range of roughly 1.3 to three times. It said most acceleration was likely due to AI assistance rather than fully autonomous contributions.
What would the proposed ban cover?
The Ban Artificial Superintelligence Act of 2026 would pause training, modification and fine-tuning of large AI systems pending rules from a proposed department. Its text also refers to recursive self-improvement and systems able to automate or greatly accelerate AI research and development.
Is the proposed ban in effect?
No. The House and Senate bills were referred to committee and had not advanced as of November 16, 2026. The source reports no change to what businesses or users can do today.
Source: ThorstenMeyerAI.com
