🔍 Read the full analysis: Could A Ban Stop Recursive Self-Improvement In AI? on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
AI systems are helping researchers develop AI, and Anthropic says Claude leads a portion of its research and development work. But the available evidence does not show an AI system independently building its own successor. A proposed congressional ban would reach beyond that scenario, though the bills have not advanced and would change nothing for companies or users now.
A viral essay by podcaster Dwarkesh Patel has renewed debate over whether AI is beginning to improve itself, as two congressional bills propose pausing some AI development and research. Evidence cited by Patel and researchers shows AI assisting with AI research, but does not document a system independently designing and building its successor; the bills remain in committee and have not changed what companies or users can do.
In an essay published November 13, Patel argued that AI is already speeding up the development of AI and forecast that progress could accelerate. His argument draws on existing assessments and company disclosures, rather than reporting a newly demonstrated system that autonomously creates its successor. The forecast is Patel’s interpretation; it should not be confused with a confirmed technical milestone.
Anthropic reported that, as of August 2026, Claude led 26% of the company’s AI research and development work. More than 90% of that work involved AI at or above a level where it collaborates, according to Anthropic’s account. The company said no measured subset of its research and development work was fully autonomous.
In a November 5 assessment based on internal data from August 12 to October 29, the nonprofit METR estimated Anthropic was improving model capabilities on well-specified tasks at about twice its rate from March 2024 to October 2025. METR described that estimate as highly uncertain, gave a possible range of 1.3 to three times, and said most of the acceleration was likely due to AI assistance. It also said AI was not yet making highly significant, fully autonomous contributions to the most important parts of model training. METR noted that Anthropic redacted much of the evidence it considered most informative.
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
What a Research Ban Could Reach
The policy debate concerns more than a hypothetical AI that builds its own successor without human involvement. The Ban Artificial Superintelligence Act of 2026 would pause training, modification and fine-tuning of large AI systems, including through recursive self-improvement, until a new Department of Artificial Intelligence sets rules. It would also prohibit systems with the capacity to automate or greatly accelerate AI research and development. That wording could reach AI-assisted research practices already in use, depending on how the law and future rules define its scope.
The distinction matters to researchers, companies and the public because AI assistance may help speed development before any system can operate independently. But evidence of faster progress is not proof that an uncontrolled self-improvement loop exists. The legislation’s broad language, combined with uncertainty about the pace and causes of progress, leaves open how a ban might affect ordinary research as well as more autonomous systems.
For now, the proposals have no immediate effect. Both bills remain at an early stage, and no law has been enacted. Claims about what the bills would mean in practice should be treated as prospective, not as a description of current restrictions.
Popular right now

Bose QuietComfort Headphones (2nd Gen), Wireless Noise-Cancelling Over-Ear Headphones, Eucalyptus Green

Dyson Airstrait Styling Brush

De'Longhi Magnifica Duo Fully Automatic Espresso Machine with Milk Frother, Silver
As an affiliate, we earn on qualifying purchases.
Two Bills, One Unsettled Definition
Recursive self-improvement, or RSI, describes a feedback loop in which an AI helps develop a more capable AI, which then helps develop the next system. Anthropic’s essay “When AI builds itself” defines the endpoint as an AI capable of fully autonomously designing and developing its own successor. The company wrote, “We are not there yet,” while cautioning that this outcome is not inevitable.
There is a broad middle ground between human-led research and fully autonomous development. Anthropic uses an automation scale that ranges from no AI involvement to full autonomy. At the “leads” level, AI completes most of a task from a high-level prompt while a person supervises. Anthropic’s reported 26% figure refers to this level of work, not to AI independently setting research goals or deciding to build a successor.
The Ban Artificial Superintelligence Act 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 latest status information in the source material lists both as referred to committee; the House bill had 12 Democratic cosponsors. A separate proposal, Rep. Ro Khanna’s Human Control Over AI Act, was described in a September CNBC report as a summary, but no bill text was available in the source material.
““We are not there yet, and recursive self-improvement is not inevitable.””
— Anthropic, in its essay “When AI builds itself”
How Much Autonomy Exists
The evidence does not settle exactly how much AI assistance is contributing to faster development across the industry. METR’s estimate covers one company, relies on internal data and carries a wide range of uncertainty; it also said Anthropic redacted much of the evidence it found most informative. The source material describes OpenAI’s figures as company-reported and says they have not received independent review.
According to The Decoder’s account of an OpenAI blog post, the company said it had met its goal for an “automated research intern” and aims for a full automated AI researcher by March 2028. The report said tasks shorter than 15 minutes succeeded 86% of the time without intervention, while more than half of successful tasks requiring four to eight hours of human work needed at least one human step. OpenAI said people still set priorities and make decisions about scaling, pauses and deployment. These company figures do not establish that AI is independently developing successor models.
It is also unclear how courts or regulators would interpret the bills’ terms, including “greatly accelerate” research, or which systems would fall within their scope. The source material gives no enacted text, implementation rules or legislative timetable.
Committee Action and New Evidence
The next immediate developments are legislative. Both Sanders’ and Casar’s bills would need to advance from committee before receiving further consideration; the source material reports no committee action or vote. The separate Khanna proposal cannot be evaluated in detail without bill text.
On the technical side, claims of recursive self-improvement will depend on clearer evidence about which research tasks AI performs, how much human supervision remains, and whether reported gains can be independently assessed. Patel’s essay has brought those questions to a wider audience, but it does not itself demonstrate that a fully autonomous loop has begun. The current record supports a narrower conclusion: AI is assisting and, in some work, leading parts of AI research while humans retain key decisions.
Key Questions
Has recursive self-improvement been demonstrated?
The cited evidence documents AI assistance and AI-led portions of research, but not an AI system fully autonomously designing and developing its own successor.
What would the proposed ban prohibit?
The Ban Artificial Superintelligence Act would pause training, modification and fine-tuning of large AI systems until a new department sets rules. Its text also targets systems able to automate or greatly accelerate AI research and development.
Are the bills law?
No. The source material lists both bills as referred to committee. They have not changed current rules for AI companies or users.
What did METR find about AI’s contribution?
METR estimated that Anthropic’s progress on well-specified tasks was about twice as fast as in an earlier period, with high uncertainty. It said AI assistance likely explained most of the acceleration, while AI was not yet making highly significant, fully autonomous contributions to key parts of model training.
Source: ThorstenMeyerAI.com
