TL;DR
Anthropic says AI assists with more than 90% of its measured AI research and development work and leads 26%, but no measured portion operates fully autonomously. A pair of congressional bills would pause some AI development and restrict systems able to automate or greatly accelerate AI research; both remain in committee.
AI is helping produce AI research, but the available evidence does not show a system autonomously building its own successor, according to Anthropic’s August 2026 measurements. The distinction matters as a viral essay by podcaster Dwarkesh Patel renews debate over rapid AI progress and two congressional bills propose restrictions that could reach systems able to automate or greatly accelerate AI research and development.
Get everyday essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
Anthropic reports that AI assists with more than 90% of its measured AI research and development work. It says Claude leads 26% of that work, meaning the system completes most of a task from a high-level prompt while a person supervises. The company says it measured no portion of the work as fully autonomous. These are Anthropic’s own figures, not an independent audit of all frontier AI labs.
Independent evaluator METR assessed that Anthropic’s progress on well-specified tasks was running at around twice its earlier rate, primarily through AI assistance rather than fully autonomous contributions. Patel’s November 13 essay draws on METR’s assessment and figures published by Anthropic and OpenAI. His further account of how quickly progress may accelerate is a forecast, not a new measurement or demonstration of an autonomous development loop.
The Ban Artificial Superintelligence Act of 2026, introduced by Sen. Bernie Sanders as S.5493 on September 23 and Rep. Greg Casar as H.R.10538 on September 24, would pause training, modification and fine-tuning of large AI systems until a new Department of Artificial Intelligence sets rules. Its text also targets systems with the capacity to automate or greatly accelerate AI research and development. The bills have not advanced beyond committee referral; the House version lists 12 Democratic cosponsors.

Where the Proposed Ban Draws Its Line
The debate affects how lawmakers might regulate AI tools used in AI development, a category that could include systems that speed up research even when people remain involved. The bill’s language reaches beyond a fully autonomous system: it also refers to the capacity to automate or greatly accelerate research and development. That makes the definition of “greatly accelerate” consequential for labs and developers.
The evidence presented so far describes substantial AI assistance at one company, alongside human supervision and no measured fully autonomous work. It does not settle how quickly other labs are progressing, or whether AI assistance has caused the measured increase in task performance. For readers, the distinction separates a documented change in research workflows from a forecast about a future system that improves itself without human direction.
Nothing in the source material indicates that the bills have changed what businesses or individual users can do today. Their practical effects would depend on whether they advance, how a new agency writes rules, and how terms such as “greatly accelerate” are interpreted.
Products Worth Considering

Samsung Galaxy Tab S10 Lite, 10.9-inch LCD, 6GB RAM, 128GB Storage, Gray

Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite

Samsung 27-Inch Odyssey OLED G6 QHD Gaming Monitor (G61SH)
As an affiliate, we earn on qualifying purchases.
From AI Assistance to a Closed Loop
Recursive self-improvement, or RSI, generally describes a loop in which an AI system helps create a more capable AI system, which then helps produce a successor. Anthropic defines the endpoint in its essay “When AI builds itself” as an AI system capable of fully autonomously designing and developing its own successor. The company says that endpoint has not been reached and is not inevitable, though it could arrive sooner than institutions expect.
Anthropic uses an automation scale attributed to Epoch AI, running from AL0, with no AI involvement, to AL5, where AI operates fully autonomously with no human in the loop. The intermediate levels distinguish AI collaboration from AI leading a task under human supervision. On Anthropic’s August 2026 account, Claude leads 26% of its AI R&D work, while no measured subset reaches full autonomy.
The renewed attention also has a political backdrop. The Associated Press called both chambers of Congress for Democrats on November 3, and supporters of an AI ban see the next Congress as a possible opportunity. A separate proposal, Rep. Ro Khanna’s “Human Control Over AI Act,” was described in a summary seen by CNBC in September. The source material says no bill text was found for that proposal.
““Recursive self-improvement is not inevitable. But it could come sooner than most institutions are prepared for.””
— Anthropic, in “When AI builds itself”
What the Measurements Leave Open
The measurements cover Anthropic’s work and are reported by the company. The source material does not establish that the same shares apply at other AI labs, or provide an independent measurement of the company’s figures. METR’s assessment concerns progress on well-specified tasks; it does not by itself show that AI assistance caused the reported rate of progress.
It is also unclear how a regulator would measure whether an AI system can “greatly accelerate” AI research, since the source material identifies no standard measure for that phrase. The bills’ potential reach would depend on rulemaking and interpretation. The proposal from Rep. Khanna cannot be evaluated from the available account because it describes a summary, not bill text.
The material does not document an AI system independently choosing research goals, designing a successor and completing its development. Patel’s forecast about faster progress should not be treated as evidence that this has happened.
Committee Action and Better Measures
The next formal milestone for S.5493 and H.R.10538 would be committee action. The latest status in the source material records referral to committee for both bills, with no further movement described. Their prospects and any changes to the text remain unknown.
For the technical debate, more comparable measurements across labs could clarify how much AI contributes to research, how researchers define task leadership, and whether those contributions change over time. For the policy debate, the definition and measurement of “greatly accelerate” will matter if lawmakers take up the bills. Until then, the evidence supports a narrower conclusion: AI is already assisting and leading some measured research tasks under human supervision, while fully autonomous recursive self-improvement has not been documented in the cited material.
Key Questions
Has recursive self-improvement been demonstrated?
The cited material does not document an AI system that fully autonomously designs and develops its own successor. Anthropic says it has measured no fully autonomous portion of its AI research and development work.
How much of Anthropic’s AI research does Claude lead?
Anthropic reports that, as of August 2026, Claude leads 26% of its AI research and development work. The company describes this as work where Claude completes most of a task from a high-level prompt while a person supervises.
What would the proposed federal ban restrict?
The Ban Artificial Superintelligence Act of 2026 would pause training, modification and fine-tuning of large AI systems until a new department sets rules. It also addresses systems able to automate or greatly accelerate AI research and development.
Have the bills become law?
No. The latest status described in the source material says both bills were referred to committee. It does not report further action.
What remains uncertain about the bill’s scope?
The source material says the bill does not define how much acceleration qualifies as “greatly,” and identifies no standard measurement for it. How regulators might apply the phrase is not yet clear.
Source: Thorsten Meyer AI
