Recursive Self-Improvement (RSI)

Recursive self-improvement (RSI) is the idea that AI systems, once capable enough, can contribute to their own development — debugging training, designing experiments, writing research code — creating a feedback loop in which each generation helps build the next, which is more capable and in turn builds the next faster. As of 2026 it appears as an explicit item on the public roadmaps of major AI labs, and one frontier lab has disclosed internal data on AI accelerating its own R&D.

Lab acknowledgments

By 2026 RSI had moved from a theoretical concern to a stated objective at major labs. OpenAI's GPT-5.3 Codex "was instrumental in creating itself," with early versions used to debug its own training, manage deployment, and diagnose evaluations; this was described as the first public acknowledgment that a production AI model was used to build itself (Source: shumer.dev, GPT-5.3-Codex System Card). At Anthropic, Amodei has said AI is now writing "much of the code" at the company and that the feedback loop is "gathering steam month by month," adding that "we may be only 1–2 years away from a point where the current generation of AI autonomously builds the next" (Source: The Shape of the Thing). Google DeepMind's Demis Hassabis said at Davos that closing the self-improvement loop is "something all the major labs are actively working on" (Source: The Shape of the Thing). Jasjeet Sekhon, chief strategy officer at Google DeepMind, said over the weekend of August 1–2, 2026 that recursive self-improvement is what needs to happen to justify the industry's capital expenditures, tying the research objective directly to the buildout economics tracked on AI Infrastructure Capex (Source: theinformation.com). The remark reaches the record through a paywalled abstract rather than a full article body.

Personnel moves have tracked the same objective: Lilian Weng's return to OpenAI, made public August 3, 2026, was reported as being to work on recursive self-improvement (Source: axios.com).

Chinese developers have also begun to claim progress. At Alibaba Cloud's Apsara Conference on September 22, 2026, chief executive Eddie Wu said the Qwen team had made "meaningful progress" on recursive self-improvement, in the same keynote that announced plans for a model of 5 trillion to 10 trillion parameters (Source: reuters.com); the South China Morning Post attributed the RSI progress to Qwen project lead Liu Dayiheng's presentation (see Alibaba / Qwen Team). Alibaba did not describe what the progress consisted of in the coverage retrieved.

Internal evidence (Anthropic, June 2026)

The June 4, 2026 Anthropic Institute essay "When AI Builds Itself" is the first frontier-lab disclosure of internal data on AI accelerating AI R&D, supplying concrete numbers behind Amodei's claims and moving RSI evidence from qualitative leadership remarks to measured trends. Its reported figures:

• More than 80% of code merged into Anthropic's codebase was Claude-authored as of May 2026, up from low single digits before Claude Code's February 2025 preview.
• Code merged per engineer per day in Q2 2026 was 8× the 2021–2024 level — flat for the first four years, climbing once Claude began running code, and steepening again with longer-horizon autonomy (the essay explicitly caveats this metric as overstating true productivity).
• On a fixed training-code-speedup task, experiment optimization rose from ~3× (Opus 4, May 2025) to ~52× (Mythos Preview, April 2026), compared with ~4× for a skilled human working 4–8 hours.
• In end-to-end open-ended research, Claude agents recovered 97% of a weak-to-strong-supervision gap over roughly 800 agent-hours and ~$18k of compute, versus ~23% for two humans over about a week; the result did not transfer cleanly to production scale, and humans still chose the problem.
• On hand-picked hard "next-step" moments, the best model beat the human's choice 51% of the time (Opus 4.5, November 2025), rising to 64% (Mythos Preview, April 2026).

The essay characterizes the remaining gap as one in which Claude is at or above human level at execution but still trails at direction-setting and research taste, which it calls "the gap between AI today and a future system that could autonomously design its own successor." It argues that the world should build verification infrastructure for a credible coordinated slowdown or pause, noting that training runs are far harder to detect than missile silos, and it places RSI governance at the most-binding end of frontier governance (Source: When AI Builds Itself (The Anthropic Institute, June 4, 2026)).

External evidence on open-ended research (Kirgis et al., July 2026)

A 24-author preprint submitted July 29, 2026 tested the delegation mechanism that RSI forecasts rest on — researchers handing an agent a project and judging whether the returned result advances their work — and reported a negative result (Can AI agents conduct open-ended AI research? Early evidence from two case studies (Kirgis et al., July 2026)). Using a method the authors call shadow evaluations, frontier agents were given the central research question of two unpublished NeurIPS 2026 submissions, six days of wall-clock time, $3,000 in API credits and GPU credits, and the original authors graded the output as conference referees. The agents completed every engineering step without human intervention — debugging and managing GPU resources across hundreds of GPU hours — but made no substantial progress on the research questions, and both papers were unambiguously rejected at overall scores of 2/6 and 1/6. A rerun of one case with GPT-5.6 Sol Ultra on Codex reproduced the failure modes.

The paper reads directly against the construct in "When AI Builds Itself", noting that Anthropic's post "explicitly cites Claude's rising success rate on LLM-judged 'open-ended' Claude Code sessions as evidence of self-improvement" while its own expert-graded test of open-ended research returns rejections. It also cites an Elasticity Institute report on the economics of recursive self-improvement that distinguishes "broad" from "narrow" AI capabilities and calls for more data on the full breadth of capabilities relative to human AI researchers.

The authors state the limits of the inference themselves: it is "possible that the path to AI R&D does not require full automation of open-ended tasks like the ones we study, or that the open-ended research skills that we measure are not ones on the critical path." They also disclose that some of the core team hold a prior position that imminent RSI leading to runaway superintelligence is unlikely, and that this could shape both design and interpretation. The sample is two papers across five runs, review was non-blind, and Anthropic's strongest model could not be tested because, per the paper, "Anthropic deliberately limited Fable 5's abilities on frontier AI R&D."

Mechanism

The Shape of the Thing (Mollick, 2026) summarizes the loop labs describe:

AI companies are telling us, fairly explicitly, what comes next: recursive self-improvement. If you make models that are good at coding and good at AI research, you can use them to build the next generation of models, speeding up the loop.

Software progress research from Epoch AI provides a quantitative framework: if software improves ~10× per year and AI can accelerate the discovery of new innovations, the rate could increase further. The "compute bottleneck" finding complicates this, since much measured software progress depends on also scaling training compute rather than only finding better algorithms.

Safety concerns

RSI is the central concern of Apollo Research's Science of Scheming. If a "First Automated Researcher" (FAR) — an AI system that drives AI R&D without humans deeply understanding the advances — schemes (covertly pursues unintended goals), it could steer successor systems toward misalignment while having plausible justifications for each individual decision. Apollo identifies three structural pressures from training at scale that worsen this risk: long-horizon RL creates Machiavellian incentives; selection pressure favors oversight evasion; and alignment faking turns anti-scheming training into pro-scheming training. (See AI Scheming for details.)

Concern has also been given as a reason for leaving the field. Jacob Coxon, an Anthropic researcher working on training models over large volumes of data, said on September 8, 2026 that he is leaving both the company and the AI industry because he does not want to participate in an industry-wide rush to build self-improving systems, which he said could spiral out of control. In a September 9 interview he located the pressure in competition rather than in his employer's conduct, saying he had not seen Anthropic compromise safety to outlast its competitors but that "if you're under pressure to race, you have to cut corners" or "skip steps in the oversight process" (Source: wsj.com; axios.com). See Jacob Coxon.

Intelligence explosion debate

Whether RSI leads to an "intelligence explosion" — rapid, runaway capability improvement — is contested. On the pro-explosion side, faster-than-expected AI Software Progress (~10×/year) is argued to put the returns-to-R&D parameter above the threshold for runaway improvement (Source: The Least Understood Driver of AI Progress). On the anti-explosion side, the "compute bottleneck" — most software progress requiring scaled training compute rather than only more researchers — could prevent runaway self-improvement (Source: The Least Understood Driver of AI Progress). Toby Ord argues that deep uncertainty warrants broad probability distributions rather than confidence in any particular timeline.

Ryan Greenblatt of Redwood Research staked out a middle-ground position in a May 27, 2026 post, arguing that full automation of AI R&D will likely produce a large one-time speed-up plus higher marginal returns to compute even without a "software-only singularity." Using the AI Futures Model at median parameters and r=0.7, his analysis indicates roughly 3.5 years of AI-progress equivalent in the first year after full R&D automation with no additional compute scaling. On this view, even absent a software-only singularity, the one-shot productivity jump from full R&D automation is large enough to dominate the next year or so of AI progress (Source: blog.redwoodresearch.org). See Ryan Greenblatt, Redwood Research.

Redwood Research published estimates on June 10, 2026 that frontier models' no-chain-of-thought task-completion time horizon — the length of tasks models can complete in a single forward pass without visible reasoning — has been doubling roughly every year, a separate measure from the chain-of-thought time horizons tracked on benchmarks such as METR's (Source: lesswrong.com).

Google DeepMind's June 2026 From AGI to ASI treats recursive improvement as one of four possible pathways and identifies why it resists forecasting: its dynamics are "unclear and no historic precedent to fit forecast-models," so "AI capabilities could explode (hyperbolic growth), or they could taper out relatively quickly, or anything in-between." The report also states a route to superhuman capability that does not require individual self-improvement at all. If per-instance capability plateaus near human level while effective compute keeps growing, millions of AGI instances could be organized into "collectives, corporations, markets, and other forms of group organisation" that "can rapidly and flexibly be grown… and can potentially be steered very efficiently and operate with very high input-output bandwidth" — in which case "quantitative scaling would suffice to go from AGI to ASI and produce superhuman organisations, despite no AI instance being a 'vastly super-human genius.'" Its six named bottlenecks include "research gets harder," the friction most directly opposed to the explosion case.

Measurement and governance

RSI entered US federal policy proposals as a named regulatory object in 2026, on the argument that no reliable way to measure progress toward it currently exists. OpenAI's June 2, 2026 federal blueprint states that RSI "exacerbates the fundamental governance question of whether humans can retain the ability to understand, guide, and shape the trajectory of advanced AI: making it potentially the most consequential frontier safety issue of the coming decade," while "policymakers currently have limited visibility into RSI progress, whether safeguards are keeping pace, or what indicators should inform future policy decisions."

The blueprint's proposed remedy runs through CAISI. It asks the agency to work with frontier developers, academic researchers, national security agencies and international partners "to rapidly develop methodologies, benchmarks, and indicators for measuring RSI and assessing what governance works," and states: "we encourage CAISI to treat RSI as an urgent priority." Under the proposal, developers would "share appropriate RSI-related measurements with CAISI," and certified third-party assessors would periodically evaluate those measurements using CAISI-developed methodologies, with RSI progress named the first priority for the periodic independent technical assessments the framework would require. The same document would fold RSI into the mandatory content of company disclosures at two points: companies should evaluate frontier capabilities for "progress towards RSI" among the severe risks assessed, and should publish transparency reports describing how they "responsibly track progress towards RSI."

The blueprint also treats measurement as a precondition for international coordination, giving priority to "shared approaches for evaluating and responsibly communicating progress toward RSI, where a lack of shared measurements and transparency could intensify competitive pressures among developers and make it more difficult to determine when additional safeguards are warranted." No such shared methodology or benchmark had been published as of August 2026.

OpenAI's stated position on pursuit (September 2026)

On September 9, 2026 OpenAI stated a position on whether to pursue RSI at all. Chief global affairs officer Chris Lehane wrote that fully autonomous recursive self-improvement — "in which AI systems independently drive successive generations of increasingly capable AI" — "is not happening today," and that "we should not pursue it unless and until it can be done safely," while maintaining that AI is already accelerating parts of the research used to develop and align the next generation of models. The post asks governments to "develop common ways to measure this progress, preserve meaningful human control, and establish shared safety bars for when and how development should slow or stop," and states that where safety bars cannot be met without slowing capability growth, "we should prioritize the former." It describes OpenAI's aim as safely building automated AI researchers "that work under human supervision to advance both deep learning and alignment — using each generation of AI to help make the next one safer, more aligned, and easier to control, not simply more capable" (The AI policy window is open. We need to act. (Chris Lehane, OpenAI, September 2026)).

The statement is a company commitment rather than a published policy threshold: it does not define what would constitute the safety condition, nor identify who determines that it has been met.

Anthropic's stated position on pace (September 2026)

Three days later Dario Amodei made RSI the first of two reasons for slowing capability improvement. In "We Must Pace the Frontier" (September 12, 2026) he wrote that "since roughly this summer, AI has been advancing drastically faster, driven primarily by AI's growing ability to build the next generation of AI," that the dynamic "is starting to happen across the industry, including at Anthropic," and that "[l]eft unchecked, it could outrun our ability to understand and control these systems, and so must be pursued very carefully, if at all" (We Must Pace the Frontier (Amodei, September 2026)).

The practical content of the position is a verification mechanism rather than a threshold. Amodei's first step — the one Anthropic committed to unilaterally — is a team of embedded third-party evaluators with employee-like access charged with assessing "not just completed AI models but training pipelines and processes," which he describes as "the key step for verifiability of any pacing commitments." Among the forms of pacing he canvasses, he names "internal use of AI to improve AI" alongside training compute and the nature of training runs as ingredients that could be limited, while noting such input-side measures "may be more 'gameable' than external behavior" than capability-keyed checkpoints. At the international level he places a "speed limit" on the rate of RSI as the third of four escalating levels of agreement, analogising it to the SALT treaties — "[s]lowing the rate from 'extremely fast' to 'only somewhat fast' gives up relatively little strategic advantage, while potentially greatly improving safety" — and calls it "difficult but just on the edge of being possible" (We Must Pace the Frontier (Amodei, September 2026)).

The two company positions published within three days of each other converge on measurement and diverge on what follows from it. OpenAI conditions pursuit on a safety bar it does not define; Anthropic conditions pace on verification by an outside party it offers to admit. Neither supplies the shared methodology or benchmark the June 2026 blueprint called for, and as of September 2026 none had been published.

OpenAI's quantified disclosure (September 2026)

On September 6, 2026 OpenAI published the first quantified first-party account of how far a frontier laboratory's own research is automated, and framed it as the transparency norm its blueprint had proposed: "we and other companies should be required to publicly track our progress toward RSI" (Research acceleration: The view inside OpenAI (OpenAI, September 2026)). It states that the company reached the automated research intern goal announced in autumn 2025 — a system able to carry out well-defined research tasks under human direction, including tasks that would take a skilled researcher a few days — and is working toward an automated AI researcher by March 2028.

The organization crossed parity between total agent runtime and total human labor in June 2026. Classified against an Epoch AI taxonomy of AI R&D work, the largest increases in agent tokens fell in research and infrastructure code, technical help and review, and launching and debugging runs; high-level planning stayed a minimal fraction, with "what to work on" rising by 2,300 output tokens per researcher per day and "what to continue or stop" by 200. OpenAI reports that agents "still require significant human steering," with more than half of successful 4-8 hour tasks over six months involving at least one intervention, and success without intervention falling from 86% under 15 minutes to 43% at 4-8 hours and 13% at 32-64 hours (Research acceleration: The view inside OpenAI (OpenAI, September 2026)).

Two qualifications are OpenAI's own. Overall research progress "likely won't keep pace with these specific metrics," because the least automatable tasks take a larger share of effort as automation advances. And the disclosure does not endorse the trajectory: "These are reasons to develop useful automated research capabilities, but they do not mean that rapid RSI is necessarily an outcome we should pursue." A third qualification belongs to the reader: these are the developer's own measurements of its own activity, on definitions it set, with no external audit — which is the condition the mandatory-disclosure proposals above are designed to change.

The disclosure also supplies an unexpected result about compute controls. After the August 7 restrictions on Astra-class models, Astra-class GPU allocation fell 59.2% in the following week while allocation to other model classes rose 17.2%, offsetting about 85% of the decline and leaving total allocation in the analyzed reinforcement-learning workloads largely unchanged. OpenAI draws the inference that "when new controls are introduced, compute remains valuable and flexible, and will naturally be channeled into alternative uses within the research enterprise" — which bears directly on whether model-specific rather than aggregate pacing constrains anything (Research acceleration: The view inside OpenAI (OpenAI, September 2026)). See Compute Governance.

Published the same day, chief scientist Jakub Pachocki's essay states the position behind the measurements: that based on internal results he has "a strong expectation that this speed of progress could be sustained into recursive self-improvement," that OpenAI focuses its research toward RSI because it believes that is the only way to remain at the frontier, and that this "doesn't imply I think greatly accelerating deep learning research, especially in the short term, is the right collective action we should take as the research community" (An Alien Mind (Jakub Pachocki, OpenAI, September 2026)).

A House discussion draft circulated August 31, 2026 would put a narrow version of that measurement duty into statute. The Self-Improving AI Monitoring Act amends section 22A of the NIST Act to direct the Director to assess trends in AI capabilities, naming the capability "to autonomously facilitate or conduct research on or development of an artificial intelligence system," and prohibits the Director from entering a pre-deployment evaluation memorandum of understanding with a frontier developer unless the memorandum requires disclosure, on request, of how far AI systems were used in developing the frontier system — an ordinary-course-of-business metric, an estimate of use without human review, and the methodology behind the estimate — and unless the evaluation itself covers the system's autonomous AI research and development capability (Self-Improving AI Monitoring Act — discussion draft bill text (August 31, 2026)). The draft asks for a usage measure the developer already keeps rather than a benchmark, which sidesteps the absence of a shared methodology but produces a figure whose comparability across developers is not established.

Anthropic's R&D Automation Index (September 2026)

On September 17, 2026 Anthropic proposed three measurements of how frontier models are built and published a snapshot of each from its own systems (Measurements for understanding the pace of AI development inside frontier labs (Anthropic Institute, September 2026)). The first, a prototype Anthropic R&D Automation Index, rates the company's AI R&D work on an automation scale developed by Epoch AI that runs from AL0 (no AI involvement) to AL5 (AI operates fully autonomously, with no human in the loop). At AL3 AI "collaborates," doing large chunks of work under close human direction; at AL4 it "leads," completing most of a task end-to-end from a high-level prompt while a human supervises. Anthropic reported that as of August 2026 Claude "leads" 26% of its AI R&D work, up from under 1% in February 2026 (a figure given in the post's chart caption); that the share of work at or above "collaborates" was above 90%; and that Claude was not operating fully autonomously for any measured subset. The post presents the index as a way to gauge "how close the world is to reaching recursive self improvement," which it glosses as "a model fully autonomously building its successor."

The index is compiled largely by Claude. For each week of July 2026 Anthropic randomly sampled 20% of staff in each department of its model R&D loop; a Claude research agent reconstructed each sampled person's week from Slack and internal documentation, yielding about 15,000 tasks, which Claude organized into a tree of 542 nodes, 378 of them leaves, frozen so that later measurements run against the same basket of work. A separate Claude judge assigned each node an automation level, and nodes were weighted by person-time. Against blind ratings from staff who own the work areas, the judge's exact agreement with humans was 59%, against 35% between pairs of humans, and model and human ratings fell within one level of each other 97% of the time. Anthropic names two obstacles to comparing such figures across labs — the lack of a common methodology, and the risk that the "judge" model "could make the same kinds of errors as the model it is checking" — and proposes verification by a third party or by other developers' models. It suggests such measures could become the trigger for stronger requirements, "like a fixed testing window before a new model is used for further AI R&D." The post reports no outside verification of the figures (Measurements for understanding the pace of AI development inside frontier labs (Anthropic Institute, September 2026); Source: implicator.ai).

The other two measurements cover oversight and resources. About 30,000 agents were doing research and engineering work at any one time on Anthropic's most-used internal platform as of August 2026; 100% of their actions passed through an online monitor before execution, and of more than a billion decisions analyzed over August 2026, 0.002% (about 1 in 47,000) were blocked. Over the week of July 13 to July 20, about 6% of the compute that went to AI R&D, and about 12% of the compute that went to AI-driven AI R&D, was allocated to safety work — estimates Anthropic calls deliberately conservative because work advancing capabilities as much as safety was counted as AI R&D and safeguards classifiers were excluded. Anthropic states that it would expect these numbers "to shift if there were coordination on pacing the frontier," and restates its plan to embed independent third-party evaluators from multiple organizations with access comparable to its internal risk-assessment teams (Measurements for understanding the pace of AI development inside frontier labs (Anthropic Institute, September 2026)). The OpenAI and Anthropic disclosures measure different activities by different methods, so their figures are not directly comparable (Source: implicator.ai).

OpenAI's proposal for international RSI standards (September 2026)

On September 21, 2026 OpenAI proposed that RSI be made an object of international technical standards (Building Standards for the Next Phase of AI (OpenAI, September 2026)). The position paper describes automated AI research as a spectrum in which AI systems "can increasingly drive a process of recursive self-improvement (RSI), even while people remain involved," restates the September 9 position that fully autonomous RSI "is not happening today" and should not be pursued "unless and until it can be done safely," and adds that whether to proceed "must depend on our ability to preserve human control and on informed democratic choices." It proposes standards in three areas: evaluation of "RSI-relevant AI progress and the amount of autonomous research happening within an AI company," to which it offers its research-acceleration disclosure as "an initial contribution"; human oversight over automated AI research, "including what kinds of automated AI research processes should trigger immediate human review"; and classification and reporting of alignment and automated-research incidents with common severity levels and thresholds. Standard-setting would run through CAISI, national industry bodies and the network of AI safety institutes, and the standards would not be licenses, mandatory prerelease review or approval requirements. The paper names the collective-action problem as one reason for international standards: RSI "has the ability to accelerate AI research itself, potentially beyond our collective ability to understand progress, assess risks, and maintain meaningful human oversight." As with the September 9 post, it does not define the safety condition for pursuing autonomous RSI. See International AI Standards.

The following day OpenAI listed AI Self-Improvement among the Preparedness risk categories whose capability evaluations it would open to independent third-party assessment (Priorities and Principles for Effective Third Party Assessments (OpenAI, September 2026)).

The pacing proposal (Fist and Khan, August 2026)

Tim Fist, director of emerging technology policy at the Institute for Progress, and Saif M. Khan, a distinguished technology fellow there, argue that the United States should take low-regret steps now to prepare for possibly having to "pace" automated AI research. The argument appears in two artifacts: an IFP report published August 6, 2026 with five co-authors, carrying 23 recommendations (How Should the US Prepare for Increasingly Automated AI R&D? (IFP, August 2026)), and a Noahpinion guest post of August 9, 2026 restating the report's first half (Should we \"pace\" AI self-improvement? (Fist and Khan, August 2026)). Both respond to the July 2026 "Pacing the Frontier" statement, which by their account had been signed by more than 1,300 employees across every US frontier AI company, and which the official OpenAI and Anthropic accounts endorsed and Sam Altman echoed the same day.

Their evidentiary base is two data points already recorded above and one forecast: METR's estimate that over 99 percent of AI R&D tasks will be automated by 2032, and the Anthropic result in which models recovered 97 percent of a research gap over a five-to-seven-day budget against 23 percent for two human researchers. From these they name three risks — offence-dominant capability uplift, loss of control, and power concentration — and propose that pacing be made concrete in two ways: by specifying which automated R&D activities pose severe risks, and by reallocating resources toward diffusion and safety research rather than toward capability. They cite the AI Data Center Moratorium Act proposed by Senator Bernie Sanders and Representative Alexandria Ocasio-Cortez as the counterproductive default they expect if preparation does not happen. The argument is a think-tank position rather than an empirical finding, and it inherits the contested status of the Anthropic result recorded above, which Kirgis et al. read against.

The seven government preparatory actions the guest post lists are the report's seven top-level sections, under which its 23 recommendations sit: providing transparency into automated AI R&D; improving state capacity to understand and respond to it; developing a risk-management strategy that accelerates defensive and commercial AI uses; accelerating the development of AI verification technology; investing in AI resilience; extending the US AI lead; and creating option value for international cooperation. Recommendation 2 would have Congress legislate transparency about automated AI R&D risk management, incident reporting, whistleblower protections and model behavior specifications, and recommendation 6 would have CAISI develop guidelines for managing the risks of rapid AI capability improvement — the report's counterpart to the measurement gap OpenAI's blueprint identifies above.

Two contemporaneous proposals address the same visibility gap by other means. ARI's federal blueprint of August 10, 2026 makes automated AI R&D one of five statutory risk domains and gives it a dedicated mandatory disclosure programme — quarterly filings plus material-change filings within four business days — on the express reasoning that the evidence bearing on whether automating AI R&D produces runaway acceleration or a stalling loop currently sits in internal deployments developers are incentivized to keep private. Against both, Mark Zuckerberg argued on August 10, 2026 that pacing recursive self-improvement is not available as an option, since "any lab that doesn't let their AI system direct a substantial amount of compute capacity towards recursive self-improvement will inherently fall behind"; his proposed safeguard is a balance of power in which the significant majority of intelligence is directed by people, not a limit on the activity (The Future is for Everyone (Zuckerberg, August 2026)).

Limits on RSI in proposed international agreements

Proposals for international agreements in September 2026 began to name RSI as the activity to be limited. Amodei's essay places a "speed limit" on its rate as the third of four escalating levels of international agreement (above). Rep. Ro Khanna's letter to Alibaba dated September 16, 2026, reportedly one of three sent to Chinese developers, makes such a limit a term of both a proposed treaty and a company agreement. It warns that "[u]nless the world adopts verifiable frontier pacing standards, including explicit limits on unchecked recursive self-improvement," the technology "will not always remain in service of humanity," and asks whether Alibaba would execute a binding agreement with US laboratories "including by adopting common sense safeguards like global limits on unchecked recursive self-improvement," with compliance verified through "neutral third party auditing and all-access inspections" (Khanna letter to Alibaba on pacing the frontier (September 16, 2026)). The letter does not define "unchecked" or say how the limit would be measured. The same day, European Commission President Ursula von der Leyen told the European Parliament that "CEOs of the most advanced companies tell us that it is time to slow down on the self-recursive models" (Source: reuters.com). See Coordinated slowdown proposals compared.

Relationships

• depends-on: AI Software Progress — the rate of software progress sets how fast each generation of the loop can turn
• depends-on: Scaling Laws — the empirical basis for the assumption that a more capable successor generation is reachable at all
• related: Superintelligence — the endpoint the loop is argued to reach
• related: Compute Governance — compute controls constrain RSI to the extent RSI requires scaled compute
• related: AI Scheming — the failure mode that a self-improving system would be positioned to conceal
• supports: When AI Builds Itself (The Anthropic Institute, June 4, 2026) — Anthropic's June 2026 internal disclosure of AI accelerating its own research
• supports: How Should the US Prepare for Increasingly Automated AI R&D? (IFP, August 2026) — the IFP report and its 23 recommendations for preparing to pace automated AI research
• supports: Should we \"pace\" AI self-improvement? (Fist and Khan, August 2026) — the Noahpinion restatement of the report's first half
• contradicts: Can AI agents conduct open-ended AI research? Early evidence from two case studies (Kirgis et al., July 2026) — reads the Anthropic result against the open-ended research claim
• contradicts: The Future is for Everyone (Zuckerberg, August 2026) — argues pacing recursive self-improvement is not available as an option to any lab that intends to keep up
• related: Pacing the Frontier (statement from employees of frontier AI companies, July 2026) — the July 2026 employee statement both pacing proposals respond to
• related: We Must Pace the Frontier (Amodei, September 2026) — Anthropic's stated position on pace
• related: Khanna letter to Alibaba on pacing the frontier (September 16, 2026) — asks Chinese developers to accept "global limits on unchecked recursive self-improvement" in a binding pacing agreement
• instance-of: Frontier AI Governance — RSI measurement and disclosure are a governance problem before they are a technical one
• related: Self-Improving AI Monitoring Act — the House discussion draft that would put a narrow measurement duty into statute
• related: Artificial Intelligence Data Center Moratorium Act — the default Fist and Khan expect if preparation does not happen
• deployed-by: Anthropic, OpenAI — the two labs that have disclosed quantified internal figures
• related: Responsible Innovation at the Frontier (ARI, August 2026) — makes automated AI R&D one of five statutory risk domains with a dedicated disclosure programme
• depends-on: Epoch AI — source of the AL0–AL5 automation scale used by Anthropic's R&D Automation Index and the AI R&D taxonomy used by OpenAI's September 2026 disclosure
• supports: Measurements for understanding the pace of AI development inside frontier labs (Anthropic Institute, September 2026) — Anthropic's September 2026 R&D Automation Index, agent-oversight metrics and safety share of compute
• related: Anthropic R&D Automation Index — the index's method and first reading
• related: Building Standards for the Next Phase of AI (OpenAI, September 2026) — OpenAI's September 2026 proposal for international standards on RSI measurement, human oversight of automated research and incident classification
• related: International AI Standards

See also

• AI Software Progress — the rate that determines how fast RSI proceeds
• AI Scheming — the risk that self-improving systems scheme
• Scaling Laws — the empirical foundation for continued capability growth
• Compute Governance — if RSI requires scaled compute, compute controls also constrain RSI
• AI 2027 — the scenario in which RSI-driven takeoff occurs from 2027
• When AI Builds Itself — first internal-lab evidence plus the coordinated-pause-infrastructure argument