Google DeepMind went looking for evidence that AI is speeding up the development of AI itself. According to the lab's own assessment, it didn't find any.

In the safety report for Gemini 4 Argon, its most powerful model, the company concluded that AI research is not moving any faster—even though AI now writes most of the code DeepMind's engineers ship. The analysis found no statistically significant acceleration in the pace of AI progress after April 1, 2026, with the lab's capability index climbing at a steady 0.1 points per month. The company describes it as its first holistic empirical assessment of ML R&D automation and acceleration.

The finding cuts against the scenario that has spooked AI executives into calling for a slowdown. It is known as recursive self-improvement: an AI builds a smarter version of itself, which builds an even smarter one, until progress outruns anyone's ability to test, regulate, or switch it off—what the report describes as "a situation where existing guardrails are severely inadequate."

"There is currently no significant end-to-end acceleration of ML R&D, AI-driven or otherwise," wrote Haydn Belfield, a Google DeepMind researcher who worked on the assessment, in a thread on X. "We therefore conclude that the risk from AI-driven ML R&D is currently low." ML R&D is shorthand for machine learning research and development—the work of building new AI models.

1/9 In the Gemini 4 Argon Safety & Security Report, we published our first holistic empirical assessment of ML R&D automation and acceleration inside Google DeepMind — combining capability benchmarks, a macro capability index, data on AI-written code and trajectory outcomes, and… pic.twitter.com/Vr8Q5Obbsp

— Haydn Belfield (@HaydnBelfield) October 14, 2026

DeepMind tracks the issue because "AI could speed up frontier capability gains faster than safety, governance, and societal adaptation can keep pace," Belfield wrote. Greater automation also means giving AI agents broader access to company systems—and more room for a misaligned model, one pursuing goals its makers never intended, to cause trouble.

A Steady Climb, Not a Spike

To measure progress, the team built GCI v1, an internal index that blends the scores of 123 AI models across 92 benchmarks into a single number per model—effectively an IQ score for AI. "Frontier Gemini capabilities have improved at a remarkably steady pace of ~0.1 GCI points/month," Belfield wrote. The distinction matters because genuine recursive self-improvement would, by definition, show up as an upward bend in exactly this kind of curve.

In the report's chart of the index, Gemini 1.5 Pro sits near 0.5 points in late 2024 and Gemini 3.8 Flash just above 2.9 in September 2026, with nearly every release hugging a straight line. Argon is the single data point poking above the trend, at roughly 3.3—though its error bar, the range in which its true score could plausibly fall, stretches low enough to touch the line.

"While Gemini 4 Argon sits above [the] trend, we trace this to increased compute and organizational investment rather than reflecting AI-driven acceleration," Belfield wrote. Google unveiled Gemini 4 Argon on September 30.

Most Code Is AI-Written, but Humans Still Steer

Most production code at DeepMind is now more than 75% AI-written, and the share of fully delegated pull requests—bundles of code changes submitted for review before they go live—rose from 7.5% to 11.9% between April and September. Even so, "~88% of pull requests still require multi-turn human steering or human code authorship—and code changes remain subject to human review before submission," Belfield wrote.

A survey of 12 senior DeepMind leaders and researchers found that automation is highest in "well-specified, verifiable execution tasks" and lowest in "open-ended scientific ideation, novel experimental design, diagnosing subtle training instabilities, and long-horizon research taste." In plain terms, AI excels at executing instructions but falls short when deciding which ideas are worth pursuing.

DeepMind also reviewed 1,389 logs of Argon working as an agent inside the company. The larger the task, the less likely the model was to finish it alone—and the more often a human stepped in. Overall, Gemini 4 Argon solved roughly half of the internal research engineering tasks it completed, far below the 90% score DeepMind treats as a red flag for full automation. That published threshold gives readers a concrete yardstick for weighing future claims of runaway automation.

A Counterpoint to Earlier Warnings

The report puts internal data behind one side of a debate that has so far been fought mainly with warnings. It landed a month after Anthropic CEO Dario Amodei argued that recursive self-improvement is already accelerating—a warning endorsed by Sam Altman and Elon Musk, and one that sent chip stocks down almost 6%. Days earlier, former Anthropic engineer Jacob Coxon had quit the industry, warning that AI could "kill us all by the end of the decade."

DeepMind's numbers cover only its own lab, and the report notes it cannot rule out that an unrestricted version of Argon crosses the company's danger thresholds for cyberattacks and bioweapons information.