A Ban On AI Self-Improvement: What Might Fall Within Its Scope?
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: A Ban On AI Self-Improvement: What Might Fall Within Its Scope? on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

TL;DR

A bill introduced in Congress would pause training, modification and fine-tuning of large AI systems, including through recursive self-improvement, until a new federal department sets rules. Current evidence documents AI assisting and leading parts of research under human oversight, not independently building its own successor. The bills remain in committee, and their scope and enforcement details are not settled.

Two bills introduced in Congress in September would pause the training, modification and fine-tuning of large AI systems, including through recursive self-improvement, until a proposed Department of Artificial Intelligence sets rules. The proposals have not advanced beyond committee referral, while the evidence cited in a recent viral debate describes AI assisting human-led research—not independently designing and building its own successor.

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 bill text would also prohibit systems with the “capacity to automate or greatly accelerate” AI research and development. The proposals would remain in effect until a new federal department established rules. As of November 16, both bills were referred to committee; the House version had 12 cosponsors, all Democrats.

The bill’s wording makes its potential reach broader than systems that independently improve themselves. A system able to greatly accelerate AI research could fall within the language, depending on how the terms are defined and applied. The source material does not establish how regulators would distinguish ordinary research assistance from covered acceleration, or which systems would meet the threshold.

The debate sharpened after podcaster Dwarkesh Patel published an essay on November 13 arguing that AI is already speeding up AI development and forecasting that progress could accelerate. His essay draws on assessments and figures published by AI companies and evaluators; its forecast is Patel’s interpretation, not a new independent finding. A separate proposal from Rep. Ro Khanna, called the Human Control Over AI Act, was described in a September CNBC report, but the source material says no bill text was available.

At a glance
reportWhen: Bills introduced September 2026; status…
The developmentA viral debate about AI self-improvement has renewed attention to the 2026 Ban Artificial Superintelligence Act, a proposal that would pause some advanced AI development while new rules are established.

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

AI collaborates or moreabove 90%
AL3 and up
AI leads, a human supervises26%
AL4
Fully autonomousno measured part
AL5 · the closed loop

Anthropic: “We are not there yet.”

The bill’s line: “the capacity to automate or greatly accelerate” AI R&DS.5493 / H.R.10538 list it as a “superintelligence precursor”. How much is “greatly” is not defined.
  1. 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.
  2. AI leads about a quarterIn 26%, Claude completes most of a task from a high-level prompt while a human supervises.
  3. No closed loopNo measured part runs fully autonomously. That is the step “recursive self-improvement” usually means.
  4. 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

The Bill Could Reach Research Tools

The proposal matters because it targets not only a hypothetical system that creates a successor without human involvement, but also AI systems able to automate or greatly accelerate AI research. That language could affect companies developing advanced models and the tools researchers use to build them. The precise reach would depend on definitions, implementation and any rules issued by the proposed department.

For now, the bills have no immediate effect on businesses or users. They have not passed, and the source material reports no change to current development practices as a result of the proposals. Their introduction nevertheless puts a policy question on the table: whether certain forms of AI-assisted research should be paused or regulated before more autonomous capabilities are demonstrated.

The debate also turns on what “self-improvement” means. Under a broad definition, AI contributes to work that advances AI. Under a narrower one, the term refers to a system independently designing and developing its own successor. Treating those as equivalent risks confusing documented human-supervised assistance with a more advanced capability that the cited material says has not been demonstrated.

Products Worth Considering

Amazon Kindle Scribe (32GB) with Premium Pen, Tungsten (Like-New)

Amazon Kindle Scribe (32GB) with Premium Pen, Tungsten (Like-New)

Samsung 57 Odyssey Neo G9 (G95NC) Dual 4K UHD 1000R Curved Gaming Monitor

Samsung 57 Odyssey Neo G9 (G95NC) Dual 4K UHD 1000R Curved Gaming Monitor

As an affiliate, we earn on qualifying purchases.

Evidence Shows Human-Led Assistance

Anthropic’s essay “When AI builds itself” defines its endpoint as “an AI system capable of fully autonomously designing and developing its own successor.” The company says, “We are not there yet.” Its figures for August 2026 say AI was involved at or above a collaboration level in more than 90% of its measured work, while Claude “leads” 26% of its AI research and development work. Anthropic also says no measured subset of that work runs fully autonomously.

Those figures use an automation scale that ranges from no AI involvement to fully autonomous operation. On the scale, “leads” means AI completes most of a task based on a high-level prompt while a person supervises. The distinction matters: AI leading a task under supervision is not the same as an AI system setting its own research goals and producing its own successor without human direction.

Independent evaluator METR assessed Anthropic’s work using internal data from August 12 to October 29. In a November 5 report, it estimated that Anthropic was improving model capabilities on well-specified tasks at around twice the rate seen between March 2024 and October 2025. METR emphasized high uncertainty, saying the true acceleration could be closer to 1.3 or 3 times, and judged that AI use likely accounted for most of the speed-up. It also said AI could not yet make 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.

““We are not there yet, and recursive self-improvement is not inevitable. But it could come sooner than most institutions are prepared for.””

— Anthropic, in “When AI builds itself”

Definitions and Enforcement Remain Open

The bills do not yet provide an established regulatory interpretation of “large AI systems” or the capacity to “greatly accelerate” AI research. The available material does not explain how the proposed department would measure that capacity, which training or modification activities would be covered, or how exceptions and enforcement would work. Those questions could determine whether the bill reaches only highly capable systems or a wider set of research tools.

The evidence also has limits. METR’s estimate carries substantial uncertainty, and the evaluator said Anthropic redacted much of the evidence it found most informative. OpenAI’s September figures were reported by The Decoder after OpenAI’s site blocked the source’s download; the account says the data were internal and had no stated independent review. These sources support claims about company research and measured tasks, not a conclusion that a system has independently built its own successor.

It is also unclear whether or when either bill will receive a committee hearing or vote. The available material describes Khanna’s proposal only through a summary seen by CNBC, not bill text, so its scope cannot be compared reliably with the Sanders-Casar bills.

Committee Action Is the Next Test

The immediate next step is congressional consideration: both Sanders-Casar bills would need to advance from committee before they could be debated or voted on by the full chambers. No hearing, vote or timetable is identified in the source material. Their current status means the proposals have not changed existing rules.

Further scrutiny would likely focus on the bill’s definitions and how a proposed Department of Artificial Intelligence would set and enforce rules. Any revised bill text, committee action or details about Khanna’s proposal could clarify whether the intended pause is limited to recursive self-improvement or also covers AI systems that substantially speed up research.

On the technical side, future evaluations may help establish whether AI contributions are becoming more autonomous, but the current cited findings remain tied to human oversight and specific task measures. Until new evidence or legislative action appears, claims that the full self-improvement loop has arrived should be distinguished from evidence that AI is already helping people develop AI.

Key Questions

Would the proposed ban cover any AI tool used in AI research?

The bill text described in the source would cover systems with the capacity to automate or greatly accelerate AI research and development, as well as training, modification and fine-tuning of large AI systems. The available material does not define the threshold or establish that every research tool would be covered.

Has an AI system built its own successor without human involvement?

The source material reports no documented example of a system independently designing and developing its successor. Anthropic says no measured subset of its AI research and development work runs fully autonomously, and METR says AI cannot yet make highly significant, fully autonomous contributions to the most important parts of model training.

Have the Sanders and Casar bills become law?

No. As of November 16, 2026, the bills were referred to committee and had not advanced, according to the source material.

What does METR’s estimate of roughly twice the rate mean?

METR estimated that, as of October 2026, Anthropic was improving model capabilities on well-specified tasks at around twice its rate from March 2024 to October 2025. METR described the estimate as highly uncertain, giving a possible range from about 1.3 to 3 times, and said AI assistance likely accounted for most of the acceleration.

Source: ThorstenMeyerAI.com

You May Also Like

FairStress: A Deeper Evaluation Of Fairness In LLM Decision-making

FairStress evaluates whether LLM decisions shift unevenly when demographic identity or persuasive challenges change, beyond the final choice alone.

How AI Companies Are Changing The Way We Watch Corporate Survival

AI firms like Firmulate are live-testing synthetic workforces, revealing how automation impacts business continuity and decision-making in real time.

Inkling: Our Open-Weights Model

Inkling has introduced an open-weights AI model aimed at advancing transparency and customization in AI research, marking a significant step forward.

How Neural Radiance Fields (NeRFs) Are Creating Photorealistic 3‑D Worlds

Forces behind NeRFs enable stunning 3D worlds from limited images, transforming virtual experiences—discover how they are revolutionizing photorealistic rendering.