🔍 Read the full analysis: A Closer Look At AI's Potential For Recursive Self-Improvement on ThorstenMeyerAI.com
Get smart everyday buys 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
Evidence cited in a viral essay points to AI systems helping researchers develop AI faster, while people still set goals and oversee the work. Anthropic says no measured part of its AI research and development runs fully autonomously, and METR describes its estimate of accelerated progress as highly uncertain. A proposed congressional ban covering recursive self-improvement has been introduced but has not advanced beyond committee referral.
AI is helping researchers develop AI systems, but the evidence available as of November 16 does not show a system autonomously designing and building its own successor. The distinction is at the center of renewed debate after podcaster Dwarkesh Patel published an essay on November 13 arguing that AI is accelerating AI development, while two congressional bills propose restrictions on systems that can automate or greatly accelerate AI research.
Anthropic reported that in August 2026, its Claude model led 26% of the company’s AI research and development work, using a scale where “leading” means completing most of a task from a high-level prompt while a human supervises. The company said more than 90% of the measured work involved AI collaboration or greater involvement. It also reported that no measured subset of its AI R&D work was fully autonomous.
The independent nonprofit METR assessed Anthropic’s progress using internal data from August 12 to October 29. In a report published November 5, it estimated that the company was improving model capabilities on well-specified tasks at around twice the rate seen from March 2024 to October 2025. METR cautioned that the estimate had high uncertainty, saying the true acceleration could be closer to 1.3 times or three times. It attributed most of the likely speed-up to AI assistance, rather than fully autonomous AI contributions.
Patel’s essay draws on METR’s assessment and figures published by AI companies. His broader predictions are his own forecast, not new experimental findings. The source article also reports that a bill introduced by Sen. Bernie Sanders and Rep. Greg Casar would pause certain AI development activities, including recursive self-improvement, until a new federal department establishes rules. The legislation has not advanced beyond committee referral.
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
Human Oversight Still Defines AI Research
The distinction between AI-assisted work and autonomous self-improvement matters because the two describe different levels of capability and oversight. AI tools can help researchers write code, test models, or complete research tasks, potentially speeding development. That does not by itself establish that a system can independently choose research goals, make a successor model, and repeat the process without people.
For policymakers and the public, the current evidence affects how claims about an imminent feedback loop should be interpreted. AI assistance is documented; fully autonomous successor-building is not. The bills’ language may nevertheless cover systems that automate or greatly accelerate research, a broader category than systems that independently build successors. Their scope could therefore affect research and development even before the strongest version of recursive self-improvement is demonstrated.
Products Worth Considering

FLEXISPOT EN1 Electric Standing Desk, 48 x 24 Inches, Black

Dell XPS Tower 8960 Desktop, Intel Core i7-13700, NVIDIA GeForce RTX 4060, 16GB DDR5, 512GB SSD

Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite
As an affiliate, we earn on qualifying purchases.
Recursive self-improvement, often shortened to RSI, generally refers to a cycle in which an AI system helps create a more capable AI system, which then helps create the next one. Anthropic has described the endpoint as an AI system able to fully autonomously design and develop its own successor. The company says that point has not been reached and is not inevitable, though it could arrive sooner than institutions are prepared for.
Anthropic’s reported figures use an automation scale ranging from no AI involvement to full autonomy. On that scale, collaboration means the AI works with people; “leading” means it does most of a task under human supervision. The company’s August figures place some of its internal work at the latter level, but do not show a measured task where researchers have left the process entirely to Claude.
METR’s report provides an outside assessment, but it relied on Anthropic’s internal data and said the company redacted much of the evidence the evaluator found most informative. Separately, The Decoder reported on OpenAI figures from a company blog post that it could not independently retrieve. According to that account, OpenAI said it had met a goal for an “automated research intern,” while people still set research priorities and make decisions about scaling, pauses, and deployment.
“We are not there yet, and recursive self-improvement is not inevitable.”
— Anthropic, in its essay “When AI Builds Itself”
Autonomy and Measurement Remain Unsettled
The available reports do not establish how much of the measured acceleration came from AI assistance, or how reliably the reported rates would apply to other companies and research tasks. METR explicitly attached high uncertainty to its estimate, and its assessment was based on internal Anthropic data, some of which was redacted. The OpenAI figures cited by The Decoder also came from the company itself and, according to the publication, had no stated independent review.
It is also unclear how the proposed legislation would be interpreted in practice. Its text covers training, modification, and fine-tuning of large AI systems, as well as systems capable of automating or greatly accelerating AI research. The source material does not establish how regulators would distinguish permitted AI assistance from prohibited acceleration if the bill were enacted. No evidence cited here shows a system completing a fully autonomous successor-development loop.
Congressional Action Remains the Next Test
The Sanders-Casar bills, S.5493 and H.R.10538, were introduced in September and remain referred to committees, according to the status records cited by the source. The House bill had 12 cosponsors, all Democrats. No change to current rules for businesses or users follows from the bills’ introduction alone.
A separate proposal, Rep. Ro Khanna’s “Human Control Over AI Act,” was described in a September CNBC report, but the source article says it found no bill text. Further legislative action, publication of fuller company evidence, or new independent evaluations could clarify both the policy proposals and the pace of AI-assisted research. Until then, the best-supported account is that AI is contributing to parts of AI development under human direction, while the fully autonomous loop remains unproven.
Key Questions
Has recursive self-improvement started?
AI is assisting AI research, but the evidence cited does not show a system autonomously building its own successor. Whether that counts as RSI depends on how broadly the term is defined.
What did Anthropic report about Claude?
Anthropic said that in August 2026 Claude led 26% of its AI research and development work. It also said no measured subset of that work was fully autonomous.
How certain is METR’s estimate of faster progress?
METR estimated that Anthropic’s progress on well-specified tasks was around twice its earlier rate, but called the estimate highly uncertain. It said the true acceleration could be closer to 1.3 times or three times.
Would the proposed ban take effect now?
No. The Sanders-Casar bills have been introduced and referred to committee, but have not advanced, according to the source material. Their introduction alone does not change current rules for businesses or users.
Source: ThorstenMeyerAI.com
