🔍 Read the full analysis: What Might A Ban On Recursive AI Improvement Include? on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Two bills introduced in September 2026 would pause certain AI training and modification, including recursive self-improvement, pending rules from a proposed federal agency. They have not advanced beyond committee referral. Current evidence describes AI assisting or leading some research with human oversight, not independently building its own successor.
Two bills introduced in Congress in September would pause some large AI systems’ training, modification and fine-tuning, including work involving recursive self-improvement, until a proposed federal agency sets rules. The measures have not advanced beyond committee referral, while a separate viral debate about AI assisting AI research has renewed attention to what the bills could cover and what current evidence shows.
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 the following day. Its text would pause training, modification and fine-tuning of large AI systems, “including through recursive self-improvement,” until a new Department of Artificial Intelligence establishes rules. The proposal also reaches systems with the “capacity to automate or greatly accelerate” AI research and development. The available legislative records show both bills referred to committee; the House bill has 12 cosponsors, all Democrats.
The bills do not establish that AI systems are already independently designing successors. Anthropic’s August 2026 figures say Claude “leads” 26% of the company’s AI research and development work, but that no measured subset runs fully autonomously. The company reports that more than 90% of this work is at or above a level where AI collaborates. Those figures describe one company’s internal work and its use of an automation scale; they are not independent measurements of the entire AI industry.
A second proposal, Rep. Ro Khanna’s “Human Control Over AI Act,” was reported by CNBC in September as a summary. The source material reports no published bill text, so its proposed scope cannot be compared in detail with the Sanders-Casar measure. Neither proposal changes requirements for businesses or users at present.
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 the Pause Could Cover
The Sanders-Casar bill’s reach may extend beyond a narrow ban on an AI system independently creating its own successor. Its language also covers training and modification of large systems and systems capable of automating or greatly accelerating AI research. That could include research workflows where people still choose goals but AI handles a growing share of the work; the exact effect would depend on how the bill’s terms are defined and applied.
For AI companies, researchers and users, the distinction matters because the proposed pause would operate before a new department sets rules. If enacted, it could affect development activity while the regulatory framework is being established. But the bill is not law, and the available information does not show that any current system meets a final legal definition of recursive self-improvement.
The debate also affects how readers interpret claims that AI development is accelerating. AI support for research is documented in company disclosures and an external evaluation, but acceleration is not the same as full autonomy. Forecasts about what may follow are not evidence that an autonomous feedback loop has already arrived.
Products Worth Considering

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

DJI Mini 5 Pro Fly More Combo Plus with DJI RC 2

Bose TV Speaker (2nd Gen) Compact Soundbar with Bluetooth and HDMI-ARC, Black
As an affiliate, we earn on qualifying purchases.
From AI Assistance to Autonomy
Recursive self-improvement, or RSI, generally describes a loop in which an AI system helps develop a more capable AI system, which then helps develop a successor. Anthropic defines the endpoint in its essay “When AI builds itself” as a system able to “fully autonomously” design and develop its own successor. The company says that endpoint has not been reached and is not inevitable.
There is a broad range between no AI involvement and full autonomy. Anthropic uses an Epoch AI scale in which AI can collaborate with people or lead a task under human supervision. Its reported August figures place Claude in the “leads” category for 26% of its AI research and development work, while researchers remain involved. These are company-reported measures, not a finding that Claude independently sets research objectives or builds a successor.
METR, a nonprofit evaluator, examined Anthropic using internal data from August 12 to October 29, 2026. In a November 5 assessment, it said Anthropic was likely improving model capabilities on well-specified tasks at around twice the rate seen from March 2024 to October 2025. METR attached high uncertainty, saying the true acceleration could be closer to 1.3 times or three times. It attributed most of the likely speed-up to AI assistance rather than fully autonomous contributions and said Anthropic redacted much of the evidence it considered most informative.
Dwarkesh Patel’s November 13 essay, “The avalanche that hit math is coming for AI development,” brought the issue wider attention. The essay draws on METR’s assessment and figures published by AI companies, then makes forecasts about future progress. Those forecasts are Patel’s analysis, not new experimental evidence. OpenAI has also published internal measures of its research automation; The Decoder reported that tasks lasting four to eight human hours often still required at least one human step when successful. The figures were not independently reviewed, according to that outlet.
“Anthropic is likely improving the capabilities of their models on well-specified tasks at around twice the rate they did between March 2024 and October 2025.”
— METR
Definitions and Evidence Still Matter
The bill’s practical scope remains uncertain because it has not been enacted or interpreted by regulators, and the proposed Department of Artificial Intelligence does not yet exist. The text’s references to large systems and to the capacity to automate or greatly accelerate AI research could cover different activities depending on how lawmakers or a future agency define them.
The evidence also has limits. Anthropic’s figures describe its own work, while METR says much of the evidence informing its assessment was redacted. OpenAI’s reported measurements are also internal, according to The Decoder, which said it found no independent review described in the company material. These sources support the conclusion that AI is assisting research; they do not establish that a system has autonomously built its own successor.
It remains unclear whether either bill will receive hearings or a vote, how the measures might change during the legislative process, and whether Khanna’s separate proposal will be introduced with full text. The political outlook may shift after the November 3 election, but no outcome or schedule is confirmed in the source material.
Committee Action and Bill Text
The next concrete milestone would be movement in the committees to which the Sanders-Casar bills were referred. A hearing, revised text or committee vote could clarify how lawmakers understand “recursive self-improvement,” which systems count as large, and what activities could be paused. No such action is confirmed in the material available for this report.
Further public details could also clarify whether Khanna’s proposal becomes a formal bill and how its provisions compare with the existing measure. Meanwhile, new evaluations and disclosures may add evidence about how much AI contributes to research and whether human oversight remains in place. Until legislation advances or new evidence changes the picture, the current status is a proposed pause—not a restriction in force—and the autonomy claims remain distinct from documented AI-assisted work.
Key Questions
Would the proposed bill ban all AI research?
The Sanders-Casar bill would pause certain training, modification and fine-tuning of large AI systems, including through recursive self-improvement, until a proposed federal department sets rules. Its text also covers systems capable of automating or greatly accelerating AI research. The precise reach would depend on definitions and implementation.
Has an AI system independently built its own successor?
The source material does not document such a case. Anthropic says no measured subset of its AI research and development work is fully autonomous, and defines the endpoint as an AI system that can independently design and develop its successor.
Are either of the bills in effect?
No. The two versions of the Ban Artificial Superintelligence Act have been referred to committee and have not advanced in the records described here. The material reports no current change for businesses or users.
How strong is the evidence that AI is speeding up AI research?
METR’s November assessment said Anthropic was likely improving model capabilities on well-specified tasks at around twice its earlier rate, but marked the estimate as highly uncertain. It said the speed-up was primarily due to AI assistance rather than fully autonomous contributions.
Source: ThorstenMeyerAI.com
