Google is “starting to get onto that RSI treadmill”, cofounder Sergey Brin said in a roundtable on Gemini 4 Argon, Business Insider reported. Google shared the video on Wednesday. Brin pointed to the Flash models Google released before Gemini 4 as evidence.
“And now our researchers and the model work hand in hand to make the next version of the model.” Sergey Brin, Google cofounder
Recursive self-improvement, or RSI, means AI helps build better AI. That speeds up the pace of new models. Brin said Google would be a couple of models ahead internally by the time people watched the video.
Brin joined Google’s Josh Woodward, Tulsee Doshi and Melvin Johnson for the conversation. Google released Argon publicly this week. Brin has been working on Gemini since his return to Google.
What Google’s safety report says
Google DeepMind published a safety and security report for Argon this month. It finds “no significant end-to-end acceleration of AI R&D, AI-driven or otherwise”. It rates the risk from AI-driven AI R&D as “currently low”.
The report says AI now automates some parts of AI research. Google surveyed 12 senior leaders and researchers for it.
“AI agents now operate as high-leverage assistants that author a majority of production code and bug fixes, though code changes remain subject to human review before submission.” Gemini 4 Argon Safety and Security Report
Most production code at Google DeepMind is now more than 75% AI-written, according to the report. The share of fully delegated pull requests rose from 7.5% in April to 11.9% in September. About 88% still need repeated human steering or code written by a human.
Humans remain key to choosing research directions, designing new experiments and diagnosing training problems, the report says. Argon solves only around half of the challenges in Google’s internal test of AI research tasks. Full automation would need a near-perfect score.
A steady pace
Google tracks the rate at which Gemini improves with an internal index built from 92 benchmarks. Before 1 April 2026, it rose by 0.091 points a month. After that date, it rose by 0.118 points a month, a difference the report calls not statistically significant.
Argon sits above that trend. Google suspects this is due to more investment and more training compute, “rather than reflecting AI-driven acceleration”.
‘A new gear’
Vihan Jain co-leads RSI and AI for science for Gemini at Google DeepMind. On X, Jain wrote that the model is “embedded in its own evolution”.
“When your post-training demand shifts from asking for raw chips to asking for Argon quota, you know the RSI flywheel has reached a new gear.” Vihan Jain, Google DeepMind
Gladstone AI interviewed researchers at Anthropic, OpenAI and Google DeepMind for a report published this week. Most believe RSI will start within three to 12 months. It could come sooner, they said, assuming no US government intervention or slowdown.
Last month, US congressman Ro Khanna said he plans a bill to ban self-improving AI. OpenAI has asked the US to lead global standards for it.