Has AI Started Improving AI? A Review Of The Evidence And Ban Debate
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Has AI Started Improving AI? A Review Of The Evidence And Ban Debate on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on office and shipping supplies

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

TL;DR

AI is helping researchers develop AI models, but available evidence does not show a system independently designing and building its own successor. A proposed federal bill would pause some AI development, including recursive self-improvement, but it remains in committee and would not change current rules.

AI is already assisting with some artificial intelligence research, but evidence reviewed so far does not show a system autonomously designing and building its own successor. The distinction is central to a renewed debate, prompted by a viral essay and a proposed congressional ban on certain AI development; the bill has not advanced beyond committee.

Anthropic reported that in August 2026, Claude led 26% of the company’s AI research and development work. More than 90% of the work measured involved AI at least collaborating, but Anthropic said no measured subset was fully autonomous. The figures describe the company’s internal work, not a general measure of the entire AI industry.

In a November 5 evaluation based on internal Anthropic data, the nonprofit METR estimated that the company was improving model capabilities on well-specified tasks at about twice its rate from March 2024 to October 2025. METR attached high uncertainty to that estimate, saying the actual rate could be closer to 1.3 or three times the earlier pace. It judged that AI assistance likely accounted for most of the acceleration, while saying AI had not made highly significant, fully autonomous contributions to the most important parts of model training.

The legislative proposal is the Ban Artificial Superintelligence Act of 2026, introduced by Sen. Bernie Sanders as S.5493 on September 23 and Rep. Greg Casar as H.R.10538 the following day. It would pause training, modification and fine-tuning of large AI systems until a new Department of Artificial Intelligence established rules. Its provisions include systems able to automate or greatly accelerate AI research and development. As of the latest status records cited in the source, both bills were referred to committee; the House bill had 12 Democratic cosponsors.

At a glance
analysisWhen: Evidence and bill status reported Novem…
The developmentA review of company disclosures, an independent evaluation and a proposed congressional bill has renewed debate over whether AI is beginning to improve itself.

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

Human Supervision Defines the Current Boundary

The distinction between AI helping researchers and AI independently improving itself matters because the two scenarios imply different levels of human control. Current evidence points to tools that can take on substantial research tasks while people set goals, supervise work and make decisions. It does not establish the autonomous feedback loop often implied by the phrase recursive self-improvement.

The debate also has immediate policy relevance, even though the proposal is not law. The bill’s wording could reach systems that accelerate research, not just a hypothetical system capable of building its own successor. If enacted, that scope could affect AI development and deployment while regulators write rules. For now, the proposal creates no new requirements for companies or users.

The evidence base has limits. METR’s assessment used internal company information, and it said Anthropic redacted much of the evidence it considered most informative. Company-reported progress and an independent evaluator’s estimate offer useful signals, but they do not settle how broadly AI is speeding up research or how long the trend will continue.

Ninja Foodi 7-Qt MegaZone FlexBasket Air Fryer, 6-in-1, Black, DZ071

Ninja Foodi 7-Qt MegaZone FlexBasket Air Fryer, 6-in-1, Black, DZ071

As an affiliate, we earn on qualifying purchases.

From AI Assistance to the Ban Proposal

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

Anthropic’s reported figures use an automation scale that distinguishes collaboration from leadership and full autonomy. Under that framework, an AI system may lead a task by completing most of it from a high-level prompt while a person supervises. That is materially different from operating with no human in the loop.

OpenAI also reported progress toward automated research. According to The Decoder, which reviewed an OpenAI post that the source article could not download, the company said it had reached its “automated research intern” goal according to its measurements and aims for a full automated AI researcher by March 2028. OpenAI’s figures, as relayed by The Decoder, said tasks lasting under 15 minutes succeeded 86% of the time without intervention. For successful tasks taking four to eight human hours, more than half required at least one human step. The Decoder noted that the figures came from OpenAI and had no stated independent review.

The topic returned to wider attention after podcaster Dwarkesh Patel published an essay on November 13 arguing that AI is already accelerating AI development and forecasting faster progress. That forecast is commentary, not a separate demonstration of autonomous self-improvement. The source article also linked renewed interest in a congressional ban to Democrats’ gains in both chambers in the November 3 elections.

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

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

Limits of the Available Evidence

How much AI caused the reported acceleration remains uncertain. METR said most of the increase was likely due to AI use but gave a wide range for the size of the effect and reported that Anthropic redacted much of the evidence behind its conclusions.

Anthropic’s and OpenAI’s figures describe work inside those companies; they do not establish the same performance across the industry. OpenAI’s reported measurements, as described by The Decoder, were not independently reviewed. It is also unclear how quickly human-supervised systems could become more autonomous, or whether they will reach the full self-improvement scenario described in the debate.

The source material identifies a separate “Human Control Over AI Act” proposal from Rep. Ro Khanna, but says it had been reported only as a summary seen by CNBC in September; no bill text was found. The scope and status of any such proposal therefore cannot be compared with the text of the Sanders-Casar bill on the same basis.

Committee Action and New Evaluations

The next formal milestone for the Sanders-Casar proposal would be committee action. The latest status cited in the source shows both bills referred to committee, with no further movement reported. Until Congress acts, the proposal does not pause AI research or change obligations for companies and users.

Further independent evaluations could clarify whether AI assistance is driving measurable gains beyond the companies’ own reports, and whether systems can complete longer research tasks with fewer human interventions. For now, the supported conclusion is narrower: AI is contributing to AI research, while fully autonomous successor-building has not been documented.

Key Questions

Has AI started improving AI?

In a limited, human-supervised sense, yes. Anthropic and METR describe AI assistance in AI research, but available evidence does not show a system independently building its own successor.

What did METR find about Anthropic’s progress?

METR estimated that Anthropic was improving model capabilities on well-specified tasks at about twice its rate from March 2024 to October 2025. It flagged high uncertainty, giving a possible range closer to 1.3 or three times the earlier rate.

Would the proposed federal bill ban all AI?

The Sanders-Casar bill would pause training, modification and fine-tuning of large AI systems pending rules from a proposed federal department. Its text also covers some systems that can automate or greatly accelerate AI research; it has not advanced beyond committee referral in the status cited.

Is a full automated AI researcher already operating?

OpenAI said it had met its automated research intern goal according to its measurements and aims for a full automated AI researcher by March 2028. The reported figures include tasks that still needed human intervention, and the source provides no independent review of OpenAI’s measurements.

Source: ThorstenMeyerAI.com

This content is for general information only and is not financial, tax or legal advice. Consult a qualified professional for decisions about your money.
You May Also Like

Can Batteries And 800-Volt DC Change AI Data Center Power?

Vantage reserved more than 1 GW of transformer capacity as operators test 800-volt DC and batteries for rising AI data center power needs.

DoorDash To Acquire Instacart

DoorDash agreed to acquire Instacart for $65 per share, valuing the company at about $17 billion. The deal still requires regulatory and shareholder approval.

Astra’s AI Repair: Editing Model Weights Without An IQ Leap

Researchers report that GPT-6 Astra raised Qwen3.5-0.8B’s MATH-500 score from 7.9% to 48.1% by editing weights, without gradient descent.

Corvus ISR: The Public Build Begins With WAMI Exploitation From Zero

Corvus ISR begins its public development of a WAMI exploitation stack, starting with synthetic data and live detection in the browser, marking a new approach in ISR software.