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

TL;DR

Anthropic reported that Claude led 26% of its AI research and development work as of August 2026, while no measured portion ran fully autonomously. A proposed federal ban would cover systems able to automate or greatly accelerate AI research, but the bills remain in committee and do not define how to measure “greatly accelerate.”

Two congressional bills would restrict AI systems that can automate or greatly accelerate AI research, as a viral essay renews debate over whether recursive self-improvement has begun. The evidence cited so far describes AI assisting and leading parts of research under human supervision; it does not document a system autonomously designing and building its own successor.

Before you orderOffer from Amazon

Get the latest gadgets 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.

Podcaster Dwarkesh Patel published an essay on November 13 arguing that AI is already speeding up the development of AI and forecasting that progress could accelerate further. His argument draws on an assessment by independent evaluator METR and figures published by frontier AI companies. The forecast is Patel’s interpretation; it is not a new measurement showing that AI systems are building their successors independently.

Anthropic’s measurements, which it says reflect work as of August 2026, put more than 90% of its AI research and development work at or above a level where AI collaborates on tasks. The company said Claude led 26% of that work, meaning it completed most of a task from a high-level prompt while a person supervised. Anthropic reported no measured subset of its AI R&D work operating fully autonomously.

METR judged that Anthropic’s progress on well-specified tasks was running at around twice its earlier rate, primarily through AI assistance rather than fully autonomous contributions. That assessment concerns the pace of progress on those tasks; it does not establish that AI has entered a self-directed cycle of successor design.

At a glance
analysisWhen: Essay published November 13, 2026; bill…
The developmentA viral essay renewed debate about recursive AI self-improvement as two congressional bills proposing restrictions remain stalled in committee.
Graphic titled Who says it, and what it rests on. Four cards. METR, highlighted, outside check: around twice the earlier pace at Anthropic, range 1.3x to 3x, primarily via AI assistance, most key evidence redacted. Anthropic, own data: Claude leads 26 percent of its AI R&D work in August 2026, no measured part fully autonomous. OpenAI, own data: automated research intern reached according to our measurements, as The Decoder reports, no independent review named. Dwarkesh Patel, commentary: essay of November 13, built on the three above; his timeline is his own forecast, not a measurement
What the claim that recursive self-improvement has started rests on. Sources: METR, Nov 5, 2026; Anthropic; The Decoder, Sep 7, 2026, on OpenAI's post; dwarkesh.com, Nov 13, 2026. Graphic: Thorsten Meyer AI.

Where the Proposed Ban Draws Its Line

The distinction matters because AI-assisted research is already reported, while fully autonomous recursive self-improvement remains undocumented in the evidence presented. A law covering systems that “greatly accelerate” research could reach some activities short of a complete, human-free loop, depending on how that phrase is interpreted.

For researchers, companies and policymakers, the boundary affects which systems could be trained or modified and when. The bills would pause certain activity until a new Department of Artificial Intelligence established rules. But the text does not specify a standard measure for “greatly accelerate,” leaving the practical scope uncertain.

The discussion also has immediate policy relevance: advocates see an opening to pursue restrictions in the next Congress after Democrats won both chambers in the November 3 election, according to the source material. The election itself does not change the status of the bills or impose new restrictions today.

As an affiliate, we earn on qualifying purchases.

From Assistance to Autonomy

Recursive self-improvement describes a feedback loop in which an AI system helps create a more capable system, which then helps develop the next one. Anthropic defines the most advanced version as an AI system that can fully autonomously design and develop its own successor. The company said that stage has not been reached and that it is not inevitable.

To describe intermediate levels, Anthropic uses an automation scale from Epoch AI, ranging from AL0, with no AI involvement, to AL5, where AI operates with no human in the loop. Between them, AI can collaborate on work or lead a task under human supervision. The company’s August figures place much of its R&D work in these intermediate categories, rather than at full autonomy.

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 source material reports both bills referred to committee, with 12 House cosponsors, all Democrats. Rep. Ro Khanna’s separate “Human Control Over AI Act” was described in a September CNBC report, but the source material says no bill text was available.

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

— Anthropic, in “When AI builds itself”

How “Greatly Accelerate” Would Be Measured

The bills’ phrase “greatly accelerate” is not defined in the material provided, and no standard measurement is identified for deciding when AI has crossed that threshold. It is therefore unclear whether the proposed restriction would cover only near-autonomous systems or also some AI-assisted research already taking place.

The Anthropic figures are the company’s own measurements of its work, not an independent audit of every frontier lab. METR’s assessment addresses well-specified tasks and does not settle whether AI has begun a broader feedback loop. The available evidence also does not establish how quickly systems might advance toward autonomous successor development.

The source material gives the bills’ latest status as referral to committee. It does not report subsequent committee action. Khanna’s proposal remains difficult to assess because the cited account describes a summary rather than published bill text.

Committee Action and New Measurements

The immediate legislative milestone is whether the House or Senate committees receiving the bills take them up. Until then, the proposals have not changed what companies or users may do under these bills. Supporters may renew efforts in the next Congress, but the source material does not identify a scheduled hearing, vote or implementation date.

On the research side, clearer comparisons would require transparent definitions of task automation, human supervision and acceleration over a stated baseline and time window. Further measurements could show whether AI’s role in research is expanding, but the figures reported here do not establish a fully autonomous self-improvement cycle.

Key Questions

Has recursive self-improvement begun?

AI is assisting with and leading some AI research under human supervision, according to Anthropic’s August 2026 measurements. The cited evidence does not document an AI system fully autonomously designing and developing its own successor.

What did Anthropic report about Claude?

Anthropic said Claude led 26% of its AI research and development work as of August 2026. It also reported that more than 90% of the work involved AI collaboration or a higher level of automation, and that no measured part ran fully autonomously.

What would the proposed federal ban restrict?

The Ban Artificial Superintelligence Act of 2026 would pause training, modification and fine-tuning of large AI systems until a new Department of Artificial Intelligence set rules. It also identifies systems with the capacity to automate or greatly accelerate AI research and development.

Have the bills become law?

No. The source material reports that S.5493 and H.R.10538 were referred to committee and had not moved further. They impose no current restrictions.

What remains unclear about the proposed ban?

The bill text does not define how much acceleration counts as “greatly,” and the source material identifies no standard measurement for applying that test. The precise reach of the restriction therefore remains uncertain.

Source: Thorsten Meyer AI

You May Also Like

Why Trust Matters When Handling Frontier Cyber AI Models

OpenAI reveals plans to restrict access to advanced cybersecurity AI models, emphasizing trust and safeguards, though specific details remain undisclosed.

Different Game, or Already Lost? Reading Mistral’s Sovereignty Bet

Mistral emphasizes European sovereignty, open weights, and local deployment to compete in AI. Is this a strategic advantage or a sign of falling behind?

Why was Kaceytron banned on Twitch? Streamer’s channel goes offline amid suspension

Streamer Kaceytron’s Twitch channel went offline amid reports of a suspension. The reasons remain unconfirmed, raising questions among viewers and fans.

The Grey Goo Nightmare: Should We Really Worry?

Keenly exploring the grey goo nightmare reveals whether this sci-fi scenario is a genuine threat or just hype—discover the truth behind nanotech safety.