AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: AI And Recursive Self-Improvement: Evidence, Uncertainty, And Bans on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the little things that make your day delivered free with Prime

  • Fast, free delivery on millions of items
  • Prime Video, Amazon Music and more included
  • Member-only deals all year
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

Evidence cited in a viral essay points to AI helping accelerate some AI research, usually with people setting goals and supervising work. Full autonomous development of successor models has not been documented. Two congressional bills would impose broad restrictions, but both remain in committee and have not changed rules for businesses or users.

A viral essay arguing that artificial intelligence is helping speed up the development of more capable AI has renewed scrutiny of both the available evidence and two proposed U.S. bills that would restrict advanced AI research. Current reports describe AI assistance in research with humans still involved; they do not document a system that independently designs and builds its own successor.

Podcaster Dwarkesh Patel published the essay “The avalanche that hit math is coming for AI development” on November 13. Its argument draws on evaluations and figures published by AI companies, while its predictions about faster future progress are Patel’s analysis, not a new experimental finding. The essay’s viral spread has brought the question of recursive self-improvement back into public discussion.

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 collaboration or greater involvement, while no measured portion was fully autonomous. Those figures describe Anthropic’s own work and classifications, not the entire AI industry.

An evaluation published November 5 by the nonprofit METR found that Anthropic was likely improving model capabilities on well-specified tasks at about twice the rate seen from March 2024 to October 2025. METR attached high uncertainty to that estimate, saying the actual acceleration could be closer to 1.3 or three times. It judged that AI use likely accounted for most of the increase, while saying AI was not yet making highly significant, fully autonomous contributions to the most important parts of model training.

The proposed restrictions are also specific but not in force. The Ban Artificial Superintelligence Act of 2026—S.5493, introduced by Sen. Bernie Sanders on September 23, and H.R.10538, introduced by Rep. Greg Casar the next day—would pause training, modification and fine-tuning of large AI systems until a new Department of Artificial Intelligence sets rules. The proposals also cover recursive self-improvement and systems able to automate or greatly accelerate AI research and development. Both bills remain referred to committee.

At a glance
reportWhen: As of November 16, 2026; the bills were…
The developmentA viral essay has renewed attention to evidence of AI-assisted AI research and two stalled congressional bills that would restrict advanced AI development.

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

AI collaborates or moreabove 90%
AL3 and up
AI leads, a human supervises26%
AL4
Fully autonomousno measured part
AL5 · the closed loop

Anthropic: “We are not there yet.”

The bill’s line: “the capacity to automate or greatly accelerate” AI R&DS.5493 / H.R.10538 list it as a “superintelligence precursor”. How much is “greatly” is not defined.
  1. 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.
  2. AI leads about a quarterIn 26%, Claude completes most of a task from a high-level prompt while a human supervises.
  3. No closed loopNo measured part runs fully autonomously. That is the step “recursive self-improvement” usually means.
  4. 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 Research Gains, Human Oversight

The debate matters because AI assistance can affect the pace of AI development even if no system is independently building its successor. If researchers can use models to complete more tasks or test ideas faster, the research cycle could speed up while people retain control over priorities, training decisions and deployment. That possibility is different from a fully autonomous feedback loop, and the distinction matters for public policy.

The bills would reach beyond a narrow ban on autonomous self-improvement. Their language would pause training and modification of large systems and restrict AI systems capable of greatly accelerating AI research. If enacted, that could affect research and development well before a system met the strongest definition of recursive self-improvement. For now, however, the proposals have not become law, and the source material says there is no immediate change for businesses or users.

Readers should also weigh the limits of the evidence. Anthropic and OpenAI report internal measures about their own work, while METR’s assessment uses internal Anthropic data. METR says much of its most informative evidence was redacted. The findings indicate a possible change in research speed, but they do not settle how broadly AI is accelerating development or whether that pace will persist.

Products Worth Considering

Sony BRAVIA XR8B 65-Inch OLED 4K HDR Smart Google TV (K-65XR8B)

Sony BRAVIA XR8B 65-Inch OLED 4K HDR Smart Google TV (K-65XR8B)

As an affiliate, we earn on qualifying purchases.

From AI Assistance to Autonomy

Recursive self-improvement, or RSI, generally describes a cycle in which an AI system helps create a more capable AI system, which then assists in creating the next one. Anthropic defines the far end of that idea as a system that can fully autonomously design and develop its own successor. The company says that point has not been reached and is not inevitable.

There is a wide gap between people doing all research work and AI operating without human involvement. Anthropic uses an automation scale in which AI may collaborate, lead a task under supervision, or operate fully autonomously. Its August figures place some work in the “lead” category, where AI completes most of a task from a high-level prompt while a person supervises. That is evidence of substantial assistance, but not evidence of independent successor development.

OpenAI has separately described progress toward an “automated research intern” and a goal of a full automated AI researcher by March 2028. The Decoder, which reviewed OpenAI’s post after its website blocked the original download, reported that short tasks were often completed without intervention, but that more than half of successful tasks lasting four to eight human hours required at least one human step. The reported measures are company figures, and The Decoder noted the absence of independent review.

A second legislative proposal, Rep. Ro Khanna’s “Human Control Over AI Act,” was reported by CNBC in September as a summary. The source material says no bill text was available. The political backdrop has also shifted: after Democrats won both chambers in the November 3 elections, supporters of AI restrictions see a possible opening in the next Congress. That does not establish that either bill will advance.

““We are not there yet, and recursive self-improvement is not inevitable.””

— Anthropic, in its essay “When AI builds itself”

Limits of the Current Evidence

The evidence does not establish that AI systems are independently designing and building successor models. Anthropic’s figures cover its own research work, and METR’s estimate concerns well-specified tasks at one company over a defined period. METR says its estimate carries high uncertainty and that Anthropic redacted much of the evidence it found most informative.

OpenAI’s reported results are also based on its internal measurements. The Decoder said it found no independent review of the figures. It is not clear how comparable the companies’ measures are, how much the results generalize beyond them, or whether AI assistance will continue to raise research productivity at the reported rate. Patel’s forecast of faster progress remains a commentary-based projection, not a confirmed outcome.

The legislative path is uncertain as well. The two Sanders-Casar bills remain in committee, and the source material gives no evidence that they have advanced toward a vote. Khanna’s proposal has only been described in a reported summary; its full language and formal legislative status are not established here. The bills’ potential penalties, including up to 20 years in prison, would depend on the final text and enactment.

Congressional and Research Milestones

The immediate next steps are further evidence and legislative action. METR’s findings can be compared with later evaluations as researchers publish more data, while company reports may clarify how much work AI completes without intervention. A key question is whether future assessments show systems taking on more of the work that humans currently direct and supervise.

In Congress, the next concrete signal would be movement on the Sanders and Casar bills from their committees, including hearings, revisions or a vote. The source material does not provide a timetable for that action. Khanna’s proposal would need published bill text before its scope could be assessed in detail. Until legislation advances and becomes law, these proposals do not change current rules for AI companies, workers or users.

Key Questions

Has recursive self-improvement begun?

AI is assisting with AI research, and some work is described as AI-led under human supervision. The available material does not document an AI system fully autonomously designing and developing its own successor.

What did METR find about the pace of AI development?

METR estimated that Anthropic was improving model capabilities on well-specified tasks at about twice the rate of the earlier period it assessed. It reported high uncertainty and said the actual rate could be closer to 1.3 or three times.

Would the proposed bills ban all AI?

The Sanders-Casar bills would pause training, modification and fine-tuning of large AI systems until a new department sets rules. They also cover systems that can automate or greatly accelerate AI research. The proposals are not in force and remain in committee.

Are AI companies or users subject to new restrictions now?

No new restrictions described in the source material have taken effect. The bills have not advanced beyond committee referral, so current rules remain unchanged based on this reporting.

What is still unknown about AI-assisted research?

It remains unclear how well company-specific results apply across the industry, whether the reported speed-up will last, and how much work AI can perform without human intervention. Some key evidence behind METR’s assessment was redacted, and the companies’ measures have not been shown to be directly comparable.

Source: ThorstenMeyerAI.com

You May Also Like

Three Public Vulnerabilities. Chained.

A chain of three publicly documented vulnerabilities enabled a supply-chain attack on TanStack npm packages on May 11, 2026, exploiting public research with no zero-day exploits.

How Much Local AI Can You Run With 16 GB On A Mac Mini?

A 16GB Mac mini can run some compact local AI models, but memory overhead, model size and speed set clear limits.

A Tour Of The Scroll-Driven Rotor In “TURBINE — WERK 9”

TURBINE — WERK 9 is an interactive web field guide built around an SVG rotor that responds to scrolling. Details of its design remain limited.

Minecraft Java Edition And SDL3: A Game-Changer For Signal Monitoring

Minecraft Java Edition now uses SDL3, a development that enhances signal monitoring for operators tracking fast-moving gaming updates.