🔍 Read the full analysis: Has AI Crossed Into Recursive Self-Improvement? What We Can Say on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
AI tools are helping researchers develop AI systems, and Anthropic and METR report signs of faster progress. But the available evidence does not show an AI autonomously designing and building its successor; a proposed U.S. ban on recursive self-improvement has not advanced beyond committee referral.
AI is helping researchers build AI, but the evidence available as of November 16 does not show a system autonomously designing and developing its own successor. The distinction is central to a renewed debate sparked by podcaster Dwarkesh Patel’s November 13 essay and a proposed congressional ban on recursive self-improvement.
Anthropic reported that in August 2026 its Claude model led 26% of the company’s AI research and development work, meaning it completed most of a task from a high-level prompt while people supervised. The company said more than 90% of the work it measured involved AI at a level of collaboration or higher, but no measured subset was fully autonomous. Anthropic says the measure comes from an automation scale developed by Epoch AI.
Independent evaluator METR assessed Anthropic using internal data from August 12 to October 29. It concluded that, as of October 2026, Anthropic was likely improving model capabilities on well-specified tasks at around twice the rate recorded from March 2024 to October 2025. METR emphasized high uncertainty: it said the actual acceleration could be closer to 1.3 times or three times that earlier rate. It judged that AI assistance likely accounted for most of the increase, while saying AI had not made highly significant, fully autonomous contributions to the most important parts of model training.
The evidence has limits. METR said Anthropic redacted much of the material it considered most informative. OpenAI has also published company-reported figures about automated research tasks, but those figures were described by The Decoder, which reviewed the post after the site blocked a direct download. OpenAI said its automated research intern milestone had been met according to its measurements; the available account says people still set research priorities and make decisions about scaling, pauses and deployment.
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
AI Assistance Is Not Autonomy
The distinction matters because AI-assisted research can speed up development without amounting to a system improving itself end to end. Researchers still choose goals, supervise work and decide whether to train or deploy models. Evidence of faster work therefore does not, by itself, establish that an autonomous self-improvement loop has begun.
The policy debate has immediate relevance even though no proposed restriction is in force. The Ban Artificial Superintelligence Act of 2026 would reach beyond a narrow definition of a model building its own successor: its text would pause training, modification and fine-tuning of large AI systems, including through recursive self-improvement, and cover systems able to automate or greatly accelerate AI research and development. That language could affect research and development well before fully autonomous systems exist, depending on how rules were written and applied.
For readers, companies and policymakers, the practical question is not only whether the most advanced form of recursive self-improvement has arrived. It is also which stages of AI research are already being automated, what humans continue to control, and whether the available measurements are independent and detailed enough to support policy choices.
Products Worth Considering

Bose Smart Ultra Dolby Atmos Soundbar

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.
The Bills and Evidence Behind the Debate
Anthropic’s essay, “When AI builds itself,” defines the endpoint as a system capable of fully autonomously designing and developing its own successor. The company says that point has not been reached and that recursive self-improvement is not inevitable. Patel’s essay, titled “The avalanche that hit math is coming for AI development,” draws on METR’s assessment and figures published by AI companies; its forecasts are commentary, not a new evaluation or independent measurement.
The Ban Artificial Superintelligence Act was introduced by Sen. Bernie Sanders as S.5493 on September 23 and by Rep. Greg Casar as H.R.10538 on September 24. The latest status records cited by ThorstenMeyerAI.com show both bills referred to committee; the House version had 12 Democratic cosponsors. A separate proposal from Rep. Ro Khanna, the Human Control Over AI Act, had been described in a September CNBC report, but the article’s review found no bill text.
The topic returned to wider attention after the Associated Press called both chambers for Democrats on November 3. Supporters of the proposed restrictions see a possible opening in the next Congress, but neither bill has advanced. The source article disclosed that Anthropic makes Claude, which assisted with research and drafting; that relationship is relevant when weighing its presentation of company findings.
“We are not there yet, and recursive self-improvement is not inevitable.”
— Anthropic, in its essay “When AI builds itself”
Limits of the Available Evidence
No documented example in the cited material shows an AI autonomously building its own successor. That does not prove such a system cannot emerge; it means the reports described here do not establish that it has. The phrase “recursive self-improvement” is also used for different levels of human involvement, from AI assisting researchers to a system operating without human direction.
METR’s acceleration estimate has high uncertainty, and the evaluator said much of its most informative evidence was redacted. OpenAI’s figures, as relayed by The Decoder, were based on internal company measurements with no independent review mentioned in that account. Anthropic’s work-share figures are also company-reported. These sources provide indications of AI-assisted research, but they do not settle how much of the speed-up is caused by AI or how well the measurements generalize beyond the companies studied.
It also remains unclear how broadly the proposed bill’s terms would apply in practice. The bills have not moved out of committee, and the separate Khanna proposal lacked bill text in the source article’s review. There is no change to business or user rules today as a result of these proposals.
Committee Action and Better Measurements
The next legislative milestone would be action in committee on the Sanders and Casar bills. Until then, their restrictions remain proposals, not law. The source material gives no scheduled committee vote or confirmed timetable for further action. Supporters may press the issue in the next Congress, but the outcome is unknown.
On the research side, clearer evidence would require more transparent reporting about which tasks AI performs, how much human supervision they require, and whether improvements carry across stages of model development. Independent evaluations could also help test company-reported claims. For now, the most defensible conclusion is narrower than the viral framing: AI is contributing to AI research, and some measures point to faster progress, but fully autonomous recursive self-improvement remains unshown.
Key Questions
Has recursive self-improvement started?
That depends on the definition. AI-assisted AI research is documented; the cited reports do not show a system autonomously designing and developing its own successor.
What did METR find about Anthropic’s progress?
METR estimated that Anthropic was improving model capabilities on well-specified tasks at around twice the rate seen from March 2024 to October 2025, as of October 2026. It described the estimate as highly uncertain and said AI assistance likely explained most of the increase.
What does Anthropic say Claude does in its research?
Anthropic reported that Claude led 26% of its AI research and development work in August 2026, under human supervision. It said no measured subset of that work was fully autonomous.
Would the proposed U.S. ban apply to AI-assisted research?
The bill text would cover training, modification and fine-tuning of large AI systems, including recursive self-improvement, and systems able to automate or greatly accelerate AI research and development. How those terms would be applied is not settled, and the bills have not advanced beyond committee referral.
Are the proposed restrictions in effect now?
No. The Sanders and Casar bills remain in committee, according to the latest status records cited in the source material. They have not changed current rules for businesses or users.
Source: ThorstenMeyerAI.com
