🔍 Read the full analysis: Recursive Self-Improvement In AI: A Question Of Evidence And Boundaries 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
A viral essay by podcaster Dwarkesh Patel has renewed debate over whether AI is beginning to improve itself. Current evidence describes AI helping researchers and leading some work under human supervision; it does not document a system autonomously designing and building its successor. A proposed federal ban remains in committee and has not changed what companies or users can do.
A November 13 essay by podcaster Dwarkesh Patel has brought renewed attention to claims that AI is speeding up AI development, but the evidence cited so far does not show a system autonomously designing and building its own successor. Anthropic’s reported figures and an assessment by independent evaluator METR describe AI-assisted research in which people still set goals and retain oversight, while a proposed congressional ban on some advanced AI development remains stalled in committee.
Patel’s essay, titled “The avalanche that hit math is coming for AI development,” draws on assessments and figures released by AI companies. Its claim that faster progress may follow is a forecast, not a newly documented demonstration of an autonomous self-improvement loop. The source article says Patel labels the forecast as his own interpretation.
Anthropic reported that, as of August 2026, Claude led 26% of the company’s AI research and development work. The company also said more than 90% of that work involved AI at or above a “collaborates” level, while no measured subset ran fully autonomously. These are company-reported figures, not an independent audit of all AI research.
METR’s November 5 assessment used internal Anthropic data from August 12 to October 29. It judged that, as of October, Anthropic was likely improving model capabilities on well-specified tasks at around twice the rate recorded between March 2024 and October 2025. METR attached high uncertainty to that estimate, saying the actual acceleration could be closer to 1.3 or 3 times. It attributed most of the speed-up to AI assistance, while saying AI was not yet making highly significant, fully autonomous contributions to the most important parts of model training. METR also said Anthropic redacted much of the evidence it considered most informative.
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 Remains Central
The distinction matters because “AI helping build AI” can describe anything from coding assistance to a system independently planning and training its successor. Treating these as the same milestone can overstate what has been demonstrated and make it harder for policymakers and readers to judge risks. The evidence cited here supports increased AI assistance and faster work on specified tasks, but not an AI research process operating without human direction.
There is also a practical policy consequence. A bill introduced in September would restrict training, modifying, or fine-tuning large AI systems until a proposed federal department established rules. Its scope includes systems able to automate or greatly accelerate AI research. If enacted, that language could reach some forms of AI-assisted development even when people remain involved. No such restrictions are in force under the proposal’s current status.
Popular right now

TCL 55-Inch T7 Series 4K QLED Smart Google TV, 55T7

Ninja Air Fryer XL, 6.5 Qt, 6-in-1, MaxCrisp, Dark Grey (AF181)

Bose QuietComfort Headphones (2nd Gen), Wireless Over-Ear Noise-Cancelling Headphones, Black
As an affiliate, we earn on qualifying purchases.
What the Proposed Ban Covers
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 on September 24. The bills would pause training, modification, and fine-tuning of large AI systems, including through recursive self-improvement, until a new Department of Artificial Intelligence established rules. The latest status cited by the source shows both measures referred to committee. The House bill had 12 Democratic cosponsors.
A separate proposal, Rep. Ro Khanna’s “Human Control Over AI Act,” was described in a September CNBC report as a summary. The source article says it found no bill text. The debate returned to wider attention after Democrats won control of both chambers in the November 3 elections, according to the Associated Press, and supporters of the ban saw an opening in the next Congress. Those political developments do not mean either bill has advanced.
Anthropic defines the endpoint of recursive self-improvement as an AI system that can fully autonomously design and develop its own successor. Its assessment says that point has not been reached and is not inevitable. For this article, Anthropic and OpenAI are central to the reporting; the source article discloses that Anthropic makes Claude, which assisted with its research and drafting.
“Anthropic is “likely improving the capabilities of their models on well-specified tasks at around twice the rate” of the earlier period it assessed.”
— METR, in its November 5 assessment
Limits of the Available Evidence
The evidence is incomplete. METR said Anthropic redacted a majority of the material it found most informative, and its acceleration estimate has a wide stated range. The findings also concern one company and specified tasks; they do not establish the pace of progress across the whole AI industry. Anthropic’s work-share figures are company-reported, while METR’s assessment relied on internal data.
OpenAI has separately said it reached an “automated research intern” milestone, according to measurements it published in September, and aims for a full automated AI researcher by March 2028. The source article relied on The Decoder’s account because OpenAI’s site blocked its download. The Decoder reported that tasks under 15 minutes succeeded 86% of the time without intervention, while more than half of successful tasks lasting four to eight human hours required at least one human step. These figures were not described as independently reviewed. It remains unclear how comparable OpenAI’s measures are with Anthropic’s or METR’s.
There is no documented example in the material provided of an AI system independently setting its research goals, producing a successor, and repeating that process without human involvement. Whether and when that capability could emerge remains uncertain.
Congress and Research Milestones
The next policy milestones are committee action on S.5493 and H.R.10538 and any publication of text for Khanna’s proposal. Until a bill advances and becomes law, the proposals do not impose new requirements on AI companies or users. The source material identifies no immediate regulatory change tied to the viral essay.
On the technical side, readers will need further evaluations that disclose methods and evidence clearly enough for independent scrutiny. OpenAI’s stated target of a full automated AI researcher by March 2028 is a company goal, not a confirmed outcome. Future reporting can compare that milestone with published measures of human intervention, task scope, and who sets research priorities. For now, the strongest supported conclusion remains narrower: AI is helping accelerate parts of AI research, while humans still direct the work described in the available assessments.
Key Questions
Has recursive self-improvement in AI begun?
That depends on the definition. AI is assisting with AI research and leading some tasks under supervision, but the cited evidence does not show a system fully 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 earlier rate it compared against. It described the estimate as highly uncertain, with a possible range closer to 1.3 or 3 times, and said AI assistance likely accounted for most of the acceleration.
Would the proposed federal ban take effect now?
No. The Sanders and Casar bills were referred to committee, according to the status cited in the source material. They have not become law and do not currently restrict AI development.
What remains unknown about AI research automation?
The available assessments do not establish how widely the findings apply across the industry, and some underlying evidence is redacted or company-reported. It is also unknown when, or whether, AI systems will be able to develop successor models without human direction.
Source: ThorstenMeyerAI.com
