TL;DR
A Nov. 13 essay by podcaster Dwarkesh Patel argued that AI is already accelerating AI research, while the degree of human oversight remains central to the debate. Anthropic’s August 2026 figures show AI leading some of its research tasks, but no measured work running fully autonomously. Two congressional bills would restrict AI development and systems able to greatly accelerate AI research; both remain in committee.
Podcaster Dwarkesh Patel’s Nov. 13 essay has renewed debate over whether AI systems are beginning to improve the process of building AI, as two bills in Congress propose limits on AI development and systems able to greatly accelerate AI research. Available figures show AI assisting with and leading some research tasks under human oversight, but no measured work at Anthropic runs fully autonomously, according to the company’s August 2026 assessment.
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Patel argues that AI is already speeding up the making of AI and forecasts that progress could accelerate. His essay draws on an assessment by the independent evaluator METR and figures published by Anthropic and OpenAI. The forecast is Patel’s interpretation of those materials, rather than new experimental evidence. The distinction matters because “recursive self-improvement” can describe a broad range of AI contributions or a narrower loop in which a system builds its own successor.
Anthropic defines that narrower endpoint as an AI system fully autonomously designing and developing its successor. In its essay “When AI builds itself,” the company said, “We are not there yet.” Anthropic’s August figures, which are the company’s own measurements, say AI is involved at the “collaborates” level or higher in more than 90% of its AI research and development work. Claude “leads” 26% of that work, meaning it completes most of a task from a high-level prompt while a human supervises. The company reported no measured subset operating fully autonomously.
METR assessed that Anthropic’s progress on well-specified tasks was running at around twice the earlier rate, “primarily via AI assistance rather than fully autonomous contributions.” That assessment supports the claim that AI assistance is contributing to faster progress on certain tasks. It does not establish that an AI system independently designs and builds its successor, or that the same rate applies across all AI research.

The Line Between Assistance and Autonomy
The disagreement has practical consequences because the proposed federal restrictions use broad language that reaches beyond a fully autonomous system. If a law covers tools that “greatly accelerate” AI research, it could apply before AI systems can independently create successors. How acceleration is measured, which systems are covered and who decides whether a threshold has been met could shape the reach of any restrictions.
For researchers and companies, the distinction between supervised assistance and autonomous development bears on which activities might be paused or regulated. For the public, it affects how policymakers weigh the potential benefits of AI-assisted research against concerns about control and the pace of development. The available evidence describes current work at one company and a METR assessment of certain tasks; it does not settle how the proposed rules would apply across the industry.
Products Worth Considering

Anker SOLIX C1000 Gen 2 Portable Power Station with 100W Solar Panel

Apple Watch Ultra 4 GPS + Cellular, 49 mm

Anker SOLIX C300 DC Portable Power Station with 60W Solar Panel, 288Wh, 300W
As an affiliate, we earn on qualifying purchases.
Two Bills, One Broad Threshold
The Ban Artificial Superintelligence Act of 2026 was introduced by Sen. Bernie Sanders as S.5493 on Sept. 23 and by Rep. Greg Casar as H.R.10538 on Sept. 24. The bills would pause training, modification and fine-tuning of large AI systems until a new Department of Artificial Intelligence sets rules. Their text includes systems with “the capacity to automate or greatly accelerate the process of artificial intelligence research and development,” as well as recursive self-improvement.
As of Nov. 16, both bills were referred to committee and had not advanced. The House measure had 12 cosponsors, all Democrats, according to the status information cited in the source material. A separate proposal, Rep. Ro Khanna’s “Human Control Over AI Act,” was described in a September CNBC report as a summary; the source material says no bill text was found. Its precise provisions therefore cannot be compared with the introduced bills.
The renewed attention followed the Nov. 3 U.S. elections, after which the Associated Press called both chambers for Democrats. Supporters of a ban see an opportunity in the next Congress, but the election result itself does not change the current status of either bill. No new restriction has taken effect under these proposals.
““It could come sooner than most institutions are prepared for.””
— Anthropic, in “When AI builds itself”
What Counts as Great Acceleration
The proposed bills do not specify in the cited text a standard measure for when AI “greatly accelerate[s]” research. The source material says no standard measurement currently answers that question. Without a defined threshold, it remains unclear which tools or research workflows could fall within the proposed restriction and how regulators would assess them.
The available automation figures also have limits: they come from Anthropic’s own measurements of its research work, with figures dated August 2026. They do not show the state of every frontier lab. METR’s cited assessment concerns well-specified tasks, and Patel’s prediction of much faster progress remains a forecast. Whether a closed loop will emerge, and when, is unknown.
Committee Action and Measurement Rules
The next formal step for S.5493 and H.R.10538 would be action in their assigned committees. As of Nov. 16, the bills had not moved beyond referral. Any further legislative progress, changes to the text or new sponsors would clarify whether Congress is moving toward a pause and how it might define covered systems.
For the wider debate, a key development would be a shared method for measuring how much AI accelerates research and how much human supervision remains. Until lawmakers define the proposed threshold and more evidence is available across labs, claims that recursive self-improvement has begun depend in part on how the term is defined. Current cited evidence shows AI-assisted and AI-led work under human supervision, not a documented system independently building its successor.
Key Questions
Has recursive AI self-improvement started?
That depends on the definition. AI assists with and leads some research tasks under human supervision, but Anthropic reported no measured work running fully autonomously or independently building a successor.
What do the proposed bills seek to restrict?
S.5493 and H.R.10538 would pause training, modification and fine-tuning of large AI systems until a new department sets rules. Their text also covers systems with the capacity to automate or greatly accelerate AI research and development.
Have either of the bills become law?
No. As of Nov. 16, 2026, both had been referred to committee and had not advanced.
What did Anthropic report about AI in its research?
For August 2026, Anthropic said AI was involved at the “collaborates” level or higher in more than 90% of its AI research and development work. Claude led 26%, with human supervision, and no measured subset ran fully autonomously.
Source: Thorsten Meyer AI
