Is AI Improving Its Own Capabilities? Evidence And Possible Rules
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Is AI Improving Its Own Capabilities? Evidence And Possible Rules on ThorstenMeyerAI.com

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.

TL;DR

Anthropic and independent evaluator METR report that AI is helping speed some AI research, with humans still setting goals and supervising work. The evidence does not show an AI system independently designing and building its successor. A proposed U.S. bill would restrict some AI development but remains in committee.

AI is helping researchers develop AI, but available evidence does not show systems independently designing and building their own successors, according to Anthropic and evaluator METR. The distinction has returned to public debate after podcaster Dwarkesh Patel published an essay on November 13 arguing that AI is accelerating AI development, while a congressional bill proposing broad limits on advanced AI research remains in committee.

Anthropic reported that in August 2026, AI was involved at the “collaborates” level or above in more than 90% of its measured AI research and development work. It said Claude “led” 26% of that work, meaning it completed most of a task from a high-level prompt under human supervision. Anthropic said no measured portion of the work was fully autonomous. The company makes Claude, which the source article disclosed helped with its research and drafting.

METR, a nonprofit evaluator, reviewed Anthropic’s internal data from August 12 to October 29. Its November 5 assessment said Anthropic was likely improving model capabilities on well-specified tasks at around twice the rate seen from March 2024 to October 2025. METR attached high uncertainty, saying the actual acceleration could be closer to 1.3 times or three times, and assessed that AI assistance likely accounted for most of the increase. It also said AI had not yet made highly significant, fully autonomous contributions to the most important parts of model training.

The evidence cited in Patel’s essay draws on those assessments and company-published figures; his predictions about much faster future progress are commentary, not new experimental findings. The source article also reported on OpenAI’s internal account of its automated research work, citing The Decoder because OpenAI’s website blocked the original post from download. The Decoder reported that OpenAI said its “automated research intern” goal had been met by its own measurements. The figures it relayed included an 86% success rate for tasks under 15 minutes without intervention, while more than half of successful tasks lasting four to eight human hours needed at least one human step.

At a glance
reportWhen: Developing; source material dated Novem…
The developmentA viral essay has renewed debate over AI improving its own capabilities, as new Anthropic and METR evidence describes faster AI-assisted research and a proposed congressional ban remains stalled.

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

Why Human Oversight Still Matters

The central issue is not whether AI already contributes to AI research; the cited company and evaluator reports say that it does. The unsettled question is whether those contributions could develop into a self-sustaining cycle in which systems take over the design and development of more capable successors. Human involvement remains a key distinction: people still choose research goals and make decisions about scaling, pausing and deployment, according to the accounts summarized in the source material.

That distinction affects how officials and the public judge both the pace of progress and possible safeguards. Faster work on defined tasks may signal a change in research productivity, but it is not by itself proof of fully autonomous self-improvement. Treating the two as equivalent could overstate what has been demonstrated; treating AI assistance as irrelevant could miss a real change in how quickly researchers can work.

For businesses and users, the reported congressional proposal does not change current rules. The source says the bills have not advanced beyond committee referral. The debate matters because the proposal would reach beyond systems already shown to build successors: it would also cover systems with the capacity to automate or greatly accelerate AI research and development.

Products Worth Considering

Bose QuietComfort Ultra Wireless Noise Cancelling Earbuds, 2nd Gen, with 20W 2-Port Wall Charger

Bose QuietComfort Ultra Wireless Noise Cancelling Earbuds, 2nd Gen, with 20W 2-Port Wall Charger

Bose QuietComfort Headphones (2nd Gen), Wireless Noise-Cancelling Over-Ear Headphones, White Smoke

Bose QuietComfort Headphones (2nd Gen), Wireless Noise-Cancelling Over-Ear Headphones, White Smoke

As an affiliate, we earn on qualifying purchases.

The Bills and the Evidence Behind Them

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. As described in the source material, it 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. It would also bar systems with the capacity to automate or greatly accelerate AI research and development.

The bills remain referred to committee, according to the latest status records cited by the source. The House measure had 12 cosponsors, all Democrats. A separate proposal from Rep. Ro Khanna, called the Human Control Over AI Act, was described in a September CNBC report as a summary; the source article said it could not find bill text. The Associated Press called both chambers of Congress for the Democrats on November 3, a result that supporters of a ban see as a possible opening in the next Congress. That political prospect is not evidence that either bill will pass.

Anthropic describes the endpoint of recursive self-improvement as an AI system that can fully autonomously design and develop its own successor. Its research scale distinguishes that endpoint from AI collaboration and AI “leading” a task under human supervision. METR’s assessment also focused on well-specified tasks and noted that Anthropic redacted much of the evidence it considered most informative. These limits matter when comparing the reported acceleration with claims about an autonomous research loop.

““We are not there yet, and recursive self-improvement is not inevitable. But it could come sooner than most institutions are prepared for.””

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

How Far Automation Has Reached

There is no documented case in the source material of an AI system independently designing and building its own successor. The available figures instead come from company reporting and an evaluator’s review of Anthropic’s internal data. METR expressed high uncertainty about the size of the reported acceleration and said many of the most informative pieces of evidence had been redacted by Anthropic.

OpenAI’s figures, as relayed by The Decoder, were also based on company measurements; the source says no independent review was cited. The exact contribution of AI to the reported productivity gains, and whether the results extend beyond particular well-specified tasks, remain unclear. Patel’s forecast that faster progress may follow should be read as his argument, not as a confirmed finding.

The final scope and prospects of the proposed restrictions are also unsettled. The bills have been referred to committees, and the source gives no evidence that Congress has scheduled further action. The text of Khanna’s separate proposal was not available to the source article.

Congress and Research Milestones

The next legislative marker is whether the two Sanders-Casar bills receive committee consideration or are reintroduced or revised in the next Congress. Until then, the source reports no change to the rules affecting businesses or users. The status of Khanna’s proposal also depends on whether formal bill text becomes available.

On the research side, future independent evaluations could clarify whether AI assistance continues to accelerate model development, how broad those gains are, and whether systems begin taking on more consequential tasks without human steps. OpenAI’s stated aim, as reported by The Decoder, is a full automated AI researcher by March 2028; that is a company goal, not a demonstrated milestone. For now, the evidence supports a narrower conclusion: AI is helping people conduct AI research, while fully autonomous development of a successor remains unshown.

Key Questions

Has recursive self-improvement begun?

AI-assisted AI research is documented in the reports described here. A system independently designing and building its own successor has not been documented in the source material.

What did METR find about Anthropic’s progress?

METR said Anthropic was likely improving model capabilities on well-specified tasks at around twice the earlier rate, comparing August–October 2026 with March 2024–October 2025. It stressed high uncertainty, giving a possible range closer to 1.3 times or three times.

Would the proposed federal bill ban all AI research?

The Ban Artificial Superintelligence Act, as described in the source, would pause training, modification and fine-tuning of large AI systems until a new department set rules. It would also target systems able to automate or greatly accelerate AI research and development. The bill has not advanced beyond committee referral.

Do the reports show AI working without human oversight?

Anthropic said Claude led 26% of its measured AI research and development work in August 2026, but no measured portion was fully autonomous. METR likewise said AI had not made highly significant, fully autonomous contributions to the most important parts of model training.

Source: ThorstenMeyerAI.com

You May Also Like

ByteDance Rejects Report That Seed AI’s Zhou Chang Is Leaving To Go Independent

ByteDance rejected a report that Seed AI research lead Zhou Chang left to start a company. His current role and plans remain unverified.

Is The 31.5″ MSI Curved FHD 250Hz Gaming Monitor Worth $129 For Sam’s Club Members?

A Slickdeals listing advertises the MSI curved FHD gaming monitor for $129 for Sam’s Club members, with free shipping for Plus members.

The Secret Factors Behind Claude Fable 5.1’S AI Index Success And Cost Line

Exploring the factors behind Claude Fable 5.1’s top AI Index score and its higher per-task cost, including model design and cost management strategies.

Why Stanford Professors’ Books Have Raised Questions About Claude’s Training

A headline alleges Anthropic used Stanford professors’ books to train Claude illegally, but the available details do not establish what happened or its legal status.