🔍 Read the full analysis: AI Recursive Self-Improvement: Signs Of Progress And Scope Of A Ban on ThorstenMeyerAI.com
Get hardware and tech essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Anthropic and an independent evaluator report that AI is helping accelerate parts of AI research, while humans still set goals and supervise work. A proposed congressional ban would pause some AI development, but the bills have not advanced and the evidence does not show an AI building its own successor autonomously.
A viral essay arguing that AI is already speeding up the development of AI has renewed attention to two congressional bills that would pause certain kinds of AI development, including recursive self-improvement. Available evidence indicates that AI is assisting and leading some research tasks under human supervision, but fully autonomous AI development of a successor model has not been documented; the bills remain in committee and would not change current rules if enacted today.
Anthropic reported that, in August 2026, more than 90% of its AI research and development work involved AI at least collaborating, and that Claude led 26% of the work. The company also said Claude was not operating fully autonomously in any measured part of its AI research and development. Anthropic defines the endpoint of recursive self-improvement as an AI system that can autonomously design and develop its own successor.
The independent evaluator METR assessed Anthropic using internal data covering August 12 to October 29. In a November 5 report, it said the company was likely improving model capabilities on well-specified tasks at around twice the rate recorded from March 2024 to October 2025. METR attached high uncertainty to that estimate, saying the true acceleration could be closer to 1.3 times or three times. It attributed most of the speed-up likely to AI assistance, while saying AI had not yet made 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.
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, modifying and fine-tuning large AI systems until a new Department of Artificial Intelligence establishes rules. Its proposed restrictions include recursive self-improvement and systems capable of automating or greatly accelerating AI research and development. Both bills were referred to committee; the House version had 12 Democratic cosponsors, according to the source 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
Why AI Research Speed Matters
The debate concerns both how quickly AI capabilities may advance and whether existing oversight would keep pace. If AI tools help researchers complete more work in less time, development could accelerate even while people retain control of research goals and deployment decisions. That is different from an AI system independently creating its successor, but it may still matter to policymakers considering how to monitor development.
The proposed bills reach beyond a narrow ban on autonomous systems. Their language would also cover systems that can automate or greatly accelerate AI research and development, potentially affecting work where humans remain involved. The distinction matters because current evidence describes a spectrum of AI assistance, not a clear switch from human-led research to autonomous self-improvement. The bills have not advanced, so their language has no present legal effect.
Products Worth Considering

Logitech MX Keys S Combo Wireless Keyboard and Mouse, Graphite

Apple MacBook Pro 16.2-inch with M5 Max, 18-core CPU, 32-core GPU, 36GB Memory, 2TB SSD, Wi-Fi 7, Space Black

LG 27GS60QC-B Ultragear 27-inch Curved Gaming Monitor (QHD, 180Hz, 1ms, 1000R)
As an affiliate, we earn on qualifying purchases.
From Research Assistance to Autonomy
Recursive self-improvement, or RSI, describes a feedback loop in which AI helps produce a more capable AI system, which could then help develop a successor. The term can refer to different levels of involvement. AI may assist with individual tasks, lead a task under supervision, or operate without a human in the loop. Anthropic uses an automation scale that places collaboration and AI-led work below full autonomy.
The November 13 essay by podcaster Dwarkesh Patel, titled “The avalanche that hit math is coming for AI development,” argues that AI is already speeding up AI development and forecasts faster progress. The essay draws on METR’s assessment and figures published by AI companies; its projection is commentary rather than new experimental evidence. Anthropic has separately said the fully autonomous endpoint has not been reached and is not inevitable, though it could arrive sooner than many institutions are prepared for.
OpenAI also published internal figures in September, according to The Decoder, which reviewed a post that the source author could not access directly. The outlet reported that OpenAI said it had reached an “automated research intern” milestone and aimed for a full automated AI researcher by March 2028. Its account said tasks lasting under 15 minutes succeeded 86% of the time without intervention, while more than half of successful tasks requiring four to eight hours of human work needed at least one human step. These are company-reported measurements, not an independent evaluation.
“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 its essay “When AI builds itself”
Limits of the Current Evidence
The evidence does not establish that an AI system has autonomously designed and built its own successor. METR’s estimate concerns progress on well-specified tasks, carries substantial uncertainty and relies on internal company data; METR said much of its most informative evidence was redacted. OpenAI’s reported measures also came from inside the company, and The Decoder noted that the company did not cite independent review.
It is also unclear how broadly the proposed legislation would apply in practice. The bill text would reach systems that can greatly accelerate AI research, but the supplied material does not describe how regulators would define that capacity or distinguish ordinary research assistance from covered activity. A separate proposal, Rep. Ro Khanna’s Human Control Over AI Act, was described in a September CNBC report as a summary; the source author found no bill text.
Anthropic and OpenAI are central to the reporting, and Anthropic makes Claude, which the source author said helped research and draft the article. That disclosure is relevant when weighing company-specific claims. The available information does not establish that the bills will pass, or when the autonomous capabilities discussed in the essay might emerge.
Congressional and Research Milestones
The next legislative step for the Sanders and Casar bills would be action in their respective committees. As of November 16, both remained referred there, with no change in law for businesses or users. The source reports that supporters see a possible opening in the next Congress following the Democrats’ victories in both chambers on November 3, but that is a political assessment, not a guarantee of legislative action.
On the technical side, further independent evaluations and clearer disclosure of company data would help establish whether AI assistance is translating into sustained acceleration, and how much human oversight remains. OpenAI’s stated target is a full automated AI researcher by March 2028; that is a company goal, not a confirmed outcome. For now, the documented development is AI-assisted and partly AI-led research, while autonomous creation of a successor remains unshown in the evidence described here.
Key Questions
Has recursive self-improvement already begun?
AI is helping with AI research, and Anthropic says Claude leads some of its research and development work. But the evidence described here does not show an AI system autonomously designing and developing its own successor.
What did METR find about the pace of AI research?
METR estimated that Anthropic was improving models on well-specified tasks at around twice its earlier rate, comparing October 2026 with March 2024 to October 2025. It described the estimate as highly uncertain, with a plausible figure closer to 1.3 times or three times.
What would the proposed Ban Artificial Superintelligence Act do?
The bill would pause training, modifying and fine-tuning large AI systems until a new Department of Artificial Intelligence sets rules. Its text includes recursive self-improvement and systems able to automate or greatly accelerate AI research and development.
Has the proposed ban taken effect?
No. As of November 16, 2026, the House and Senate bills had been referred to committee and had not advanced. They do not change current rules for companies or users.
When might AI become a fully automated researcher?
OpenAI reportedly set a goal of reaching a full automated AI researcher by March 2028, according to figures reviewed by The Decoder. That is a company target, not a confirmed forecast, and the available evidence does not establish when or whether it will be reached.
Source: ThorstenMeyerAI.com
