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

🔍 Read the full analysis: AI Recursive Self-Improvement: How Far Has It Actually Progressed? 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

AI is assisting and, in some cases, leading parts of AI research, but available evidence does not show systems autonomously designing and building their own successors. Anthropic and METR report faster progress, with substantial uncertainty about the scale and causes; proposed legislation to restrict the development loop has not advanced beyond committee referral.

AI is helping researchers build AI systems faster, but evidence available as of November 16 does not show an AI system independently designing and developing its own successor. The distinction matters as a viral essay by podcaster Dwarkesh Patel argues that AI-driven research could accelerate sharply, while bills in Congress propose restrictions that include recursive self-improvement.

Anthropic reported that, in August 2026, its Claude models led 26% of the company’s AI research and development work, using a scale in which AI leads a task but a human supervises. The company said more than 90% of the measured work involved AI at the level of collaboration or above. It also said no measured subset of its AI research and development was fully autonomous. Those are company-reported figures, not evidence that a system has independently built its replacement.

The independent evaluator METR assessed Anthropic using internal data from August 12 to October 29. In a November 5 report, it said Anthropic was likely improving model capabilities on well-specified tasks at around twice the rate recorded from March 2024 to October 2025. METR stressed that estimate had high uncertainty: it said the true acceleration could be closer to 1.3 or three times. The evaluator judged that AI assistance likely accounted for most of the speed-up, while saying AI had not yet made highly significant, fully autonomous contributions to the most important parts of model training.

Patel’s November 13 essay, “The avalanche that hit math is coming for AI development,” draws on METR’s assessment and figures published by AI companies, then offers a forecast about what may follow. The essay’s forecast is commentary, not a new experimental finding. A separate policy development is more concrete: the Ban Artificial Superintelligence Act of 2026, introduced by Sen. Bernie Sanders and Rep. Greg Casar in September, would restrict certain AI training and development activities. Both bills remain referred to committee.

At a glance
reportWhen: Developing; the cited assessments and l…
The developmentA viral essay renewed debate over AI recursively improving itself, as published evaluations show AI assistance speeding research but not full autonomous successor-building.

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 AI-Assisted Research Matters

AI systems that help researchers write code, test ideas, or complete technical tasks can increase the pace of AI development even when people still set priorities and make decisions. If that assistance grows, it could affect how quickly companies release models and how much oversight regulators and the public can apply. But faster research is not the same as autonomous self-improvement: the degree of human involvement is central to both the evidence and the policy debate.

The distinction also affects proposals aimed at controlling the loop. Sanders and Casar’s bill would cover more than a system independently creating its successor. Its text would pause training, modification, and fine-tuning of large AI systems until a new Department of Artificial Intelligence established rules, and would bar systems with the capacity to automate or greatly accelerate AI research and development. That scope could reach tools used in research even when humans remain involved. No changes for businesses or users follow from the bills today, because neither has advanced beyond committee referral.

There are limits to what the evidence can establish. METR said Anthropic redacted much of the evidence it considered most informative, and it reported a broad range for its acceleration estimate. Company-published measurements can describe internal use, but they are not the same as independently verified demonstrations of autonomous research. Readers should treat claims about the pace of progress with attention to who measured it, what tasks were included, and how much human intervention remained.

Products Worth Considering

LG 27GS60QC-B Ultragear 27-inch Curved Gaming Monitor (QHD, 180Hz, 1ms, 1000R)

LG 27GS60QC-B Ultragear 27-inch Curved Gaming Monitor (QHD, 180Hz, 1ms, 1000R)

Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite

Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite

As an affiliate, we earn on qualifying purchases.

From Human Guidance to Autonomy

Recursive self-improvement is commonly used to describe a feedback loop in which an AI system helps create a more capable AI, which then helps build a later system. Anthropic defines the stronger endpoint as an AI able to fully autonomously design and develop its own successor. The company has said that endpoint has not been reached and is not inevitable.

Between ordinary human-led work and full autonomy are several levels of assistance. Anthropic uses an automation scale ranging from no AI involvement to fully autonomous operation. Its “collaborates” and “leads” categories describe substantial AI contributions while people remain in the process; “leads” means a model can complete most of a task from a high-level prompt under human supervision. The August figures therefore point to AI involvement in current research, not a human-free development loop.

OpenAI also reported progress toward automating research tasks in a September blog post. The Decoder, which reviewed the post after it was unavailable for direct download, reported that OpenAI said it had reached its “automated research intern” goal and was aiming for a full automated AI researcher by March 2028. According to The Decoder’s account, tasks lasting 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. OpenAI said people still choose research priorities and make decisions about scaling, pauses, and deployment. The reported figures are company measurements without independent review cited by The Decoder.

The legislation surfaced alongside the public debate but has not become law. Sanders introduced S.5493 on September 23, and Casar introduced H.R.10538 on September 24. The latest status described in the source material shows both referred to committee; the House bill had 12 Democratic cosponsors. Rep. Ro Khanna’s separate “Human Control Over AI Act” was reported by CNBC as a proposal summary, but no bill text was identified in the source material.

“As of October 2026, Anthropic is likely improving the capabilities of their models on well-specified tasks at around twice the rate they did between March 2024 and October 2025.”

— METR

Limits of the Available Evidence

Whether AI assistance is accelerating across the field, and by how much, remains uncertain. METR’s estimate concerns Anthropic and specified tasks, relies on internal data, and carries a broad uncertainty range. The evaluator also said Anthropic redacted a majority of the evidence it found most informative. Anthropic’s work-share figures and OpenAI’s reported milestones are company measurements; the material available here does not establish independent verification of all those figures.

It is also unclear how far current systems could progress if researchers gave them more responsibility, or whether the reported gains will continue. No cited assessment documents an AI system autonomously developing its own successor. Patel’s argument about a possible future acceleration is a forecast, and the available information does not show when, or whether, that scenario would occur.

The legislative outcome is unsettled as well. The two Sanders-Casar bills have not moved beyond committee referral, and the available source material does not establish what provisions lawmakers might retain or whether either proposal will advance. The text and status of Khanna’s proposal remain less clear because the source identifies a summary but no bill text.

Research and Bills to Watch

The next evidence to watch is whether independent evaluators publish further measurements of AI’s role in research, with clearer methods and more information about human intervention. Company disclosures may add detail, but the key test is whether systems can take on important research work without people setting direction or stepping in. Any claims of autonomous successor-building would need to be distinguished from models that assist researchers or lead bounded tasks under supervision.

On the policy side, the next formal milestones would be committee action on S.5493 or H.R.10538, or the introduction of full text for any related proposal. Until then, the bills do not impose restrictions on businesses or users. For now, the evidence supports a measured conclusion: AI is contributing to and may be accelerating AI development, but the fully autonomous loop remains unshown.

Key Questions

Has recursive self-improvement started?

AI is already helping with AI research, and Anthropic reports that Claude leads some of its research and development work under human supervision. The evidence cited here does not show an AI system independently designing and developing its own successor.

What did METR find about the pace of AI research?

METR estimated that Anthropic was improving model capabilities on well-specified tasks at around twice its rate from March 2024 to October 2025. It described the estimate as highly uncertain, with a possible range closer to 1.3 to three times, and said AI assistance likely accounted for most of the speed-up.

Do the proposed US bills ban AI development now?

No. The Sanders and Casar bills have been introduced and referred to committee, but the source material reports no further progress. They are proposals, not current restrictions.

What is the difference between AI assistance and autonomous self-improvement?

AI assistance means a model helps with tasks while people direct the work or supervise it. Autonomous self-improvement, as defined by Anthropic, would mean a system can design and develop its own successor without human involvement; that has not been documented in the cited evidence.

Source: ThorstenMeyerAI.com

You May Also Like

Find A 3D Printer For Your Next Project: 10 Picks For 2027

A 10-printer comparison highlights tradeoffs in enclosure, multicolor printing, speed and beginner-friendly size. The source does not confirm 2027 model updates.

The Claude Code Team At Anthropic Sees Three People Depart

Analytics India Magazine reported three departures linked to Claude Code within 10 days. Their identities, roles and reasons have not been disclosed.

10 Best Mobile Workstations For AI-Focused Work In 2027

A source-based comparison of 10 mobile workstation configurations, with attention to GPU, memory, portability and limits in the available specifications.

Inside The Tech Of AI-Driven Interactivity At ‘Lot 87 — The Varos Evening Sale’

Exploring how AI-driven interactivity transformed ‘Lot 87’ into an immersive, tech-powered auction experience with custom scripts and dynamic elements.