🔍 Read the full analysis: AI Self-Improvement And Bans: What The Current Evidence Shows on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
AI systems are helping researchers develop AI, but the evidence cited so far does not show a system autonomously designing and building its own successor. Two congressional bills would restrict AI development, including recursive self-improvement, but both remain in committee and have not changed current rules.
A viral essay about AI speeding up its own development has renewed debate over recursive self-improvement, but the evidence currently cited shows AI assisting human researchers—not systems independently building their successors. Two bills in Congress would restrict AI development, including this kind of research loop, yet both remain in committee and have not changed what companies can do today.
In a November 13 essay and narrated episode, podcaster Dwarkesh Patel argued that AI is already accelerating the work of making AI and forecast that progress could speed up further. The essay draws on existing assessments and figures published by AI companies; its forecast is commentary, not new empirical evidence. The distinction matters because “self-improvement” can describe anything from AI helping with research to a system independently designing its successor.
Anthropic reported that in August 2026, Claude led 26% of the company’s AI research and development work. The company also said that more than 90% of the work it measured involved AI at or above a level where it collaborates, but that no measured subset was fully autonomous. These figures are company-reported measures, not evidence that Claude can run the full research process alone.
An independent nonprofit evaluator, METR, assessed Anthropic using internal data from August 12 to October 29. In a November 5 report, METR said Anthropic was likely improving models’ capabilities on well-specified tasks at around twice the rate seen from March 2024 to October 2025. It stressed that this estimate had high uncertainty: METR said the actual acceleration could be closer to 1.3 times or three times. It judged that AI assistance likely accounts for most of the speed-up, while saying AI could not yet make highly significant, fully autonomous contributions to the most important parts of model training.
METR also disclosed that Anthropic redacted much of the evidence it considered most informative. Its findings therefore offer an outside assessment based on access to company data, but readers cannot inspect all the underlying material.
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 AI-Assisted Research Changes
The distinction between assistance and autonomy shapes both the safety debate and the proposed legislation. AI tools that help researchers write code, analyze results or complete bounded tasks can speed up development while people still set research goals and make decisions. A system that independently plans, conducts and applies research would represent a different level of capability and control.
METR’s estimate, if broadly accurate, suggests AI may already be contributing to faster progress on certain tasks. But the report does not establish that AI alone caused the change, and it does not show that a system can take over the full process. The figures point to a meaningful question for oversight—how much of AI development is being automated—without confirming the stronger claim that AI has become self-improving in the fully autonomous sense.
The policy stakes are immediate even though the bills have not advanced. The proposed restrictions reach beyond a narrow ban on autonomous systems: they could cover training, modification and fine-tuning of large AI systems, as well as systems capable of greatly accelerating AI research. That breadth could affect companies, researchers and users if legislation were enacted. No such change is in force under these proposals now.
Products Worth Considering

Logitech MX Master 4 for Mac Wireless Ergonomic Mouse, White Silver

Vari Ergo Electric Standing Desk, 54×26, Light Wood/White

FLEXISPOT EN1 Electric Standing Desk, 48 x 24 in, Black
As an affiliate, we earn on qualifying purchases.
How the Bills Define the Risk
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. The bills propose pausing the training, modification and fine-tuning of large AI systems—including through “recursive self-improvement”—until a new Department of Artificial Intelligence establishes rules. They would also prohibit systems with the capacity to automate or greatly accelerate AI research and development.
As of the source’s November 16 account, both bills had been referred to committee and had not advanced. The House version had 12 Democratic cosponsors. A separate proposal, Rep. Ro Khanna’s “Human Control Over AI Act,” was described in a September CNBC report, but the source found no bill text to assess. The Associated Press reported that Democrats won control of both chambers in the November 3 elections; supporters of a ban see that result as a potential opening in the next Congress, not as a change to current law.
Anthropic describes the endpoint of recursive self-improvement as an AI system fully autonomously designing and developing its own successor. Its framework distinguishes levels of involvement: AI can collaborate with researchers or lead much of a task under human supervision. Those intermediate levels are not the same as a system operating without a human in the loop.
“We are not there yet, and recursive self-improvement is not inevitable.”
— Anthropic, in its essay “When AI Builds Itself”
Limits in the Available Evidence
The available material does not show an AI system independently designing and building its own successor. That conclusion is limited to the evidence described in the source; it is not proof that such a capability cannot emerge or that no unreported work exists. Anthropic said its measurements found no fully autonomous subset of its AI research work, while METR described its own acceleration estimate as highly uncertain and noted that much of the relevant evidence had been redacted.
Other figures also rely on company reporting. OpenAI published internal measurements in September, but the source relied on The Decoder’s account because it could not download OpenAI’s post. The Decoder reported that OpenAI said it had met its “automated research intern” goal according to its own measurements. It also reported that tasks lasting under 15 minutes succeeded 86% of the time without intervention, while more than half of successful tasks taking four to eight human hours required at least one human step. The account noted no independent review of those company data.
It remains unclear how much of the reported research acceleration is caused by AI, how well internal measurements generalize outside company settings, and how the proposed bills’ broad terms would be defined in practice. The proposals’ eventual scope and prospects also depend on committee action and any later revisions.
Committee Action and New Measurements
The next legislative milestone is whether the committees handling S.5493 and H.R.10538 take them up. Until then, the bills remain proposals, and the source reports no immediate change for businesses or users. The status of Khanna’s proposal is harder to track because the source found only a summary, not legislative text.
On the evidence side, future independent assessments could clarify whether AI’s role in research is growing and whether companies’ internal measures match outside evaluations. Readers should distinguish updates to task-level performance and AI assistance from evidence of full autonomy. The central question remains whether human-supervised acceleration progresses into systems that can direct and complete AI development on their own.
Key Questions
Has recursive AI self-improvement been demonstrated?
The evidence described so far shows AI assisting and sometimes leading parts of research under human supervision. It does not show a system fully autonomously designing and developing its own successor.
What did METR say about the pace of Anthropic’s progress?
METR estimated that Anthropic was improving models on well-specified tasks at around twice the rate measured between March 2024 and October 2025. It called the estimate highly uncertain and said the true figure could be closer to 1.3 times or three times.
Would the proposed bills ban all AI research?
The bills propose pausing training, modification and fine-tuning of large AI systems until a new department sets rules. They also cover systems capable of automating or greatly accelerating AI research. Their practical reach would depend on the bill text and later implementation; neither proposal has become law.
Are the congressional proposals currently in effect?
No. As of November 16, 2026, the Sanders and Casar bills had been referred to committee and had not advanced. The source reports no current restrictions resulting from these proposals.
Source: ThorstenMeyerAI.com
