Brin: Gemini Nears Self-Improvement Milestone as AI Helps Build Next Models

Google co-founder Sergey Brin said at a roundtable discussion around the new Gemini 4 Argon model that Gemini is approaching a crucial "self-improvement" moment, with AI already beginning to help build its next-generation models.

Brin noted that Google's Flash models have already shown this trend, and the company is "starting to step on the RSI treadmill." RSI, or Recursive Self-Improvement, refers to AI helping build stronger AI, creating a flywheel effect of accelerating iteration.

In Google's Argon safety report, the company disclosed that parts of the AI development workflow have been automated: AI agents can now handle most production code writing and vulnerability fixes, though code changes still require human review before submission.

This view is not unique to Google. Anthropic and OpenAI have also said AI may be nearing the RSI stage. A Gladstone AI report published in October 2026 said most surveyed researchers from institutions including Anthropic, OpenAI, and DeepMind believe RSI could begin within 3 to 12 months.

However, RSI has also raised safety concerns. AI researchers have previously warned that without sufficient safeguards, humans could gradually lose control over rapidly self-improving technology.