# Demis Hassabis Wants an AI Regulator With Teeth — And a 30-Day Rule > Source: > Published: 2026-07-14 13:12:08+00:00 # Demis Hassabis Wants an AI Regulator With Teeth — And a 30-Day Rule Google DeepMind CEO Demis Hassabis proposed a FINRA-style self-regulatory body for frontier AI models on July 14, calling for mandatory 30-day pre-release safety assessments. The framework would apply to all models regardless of origin or open/closed status, and includes government authority to mandate development slowdowns. #### Google [DeepMind](/glossary/deepmind) CEO Demis Hassabis proposed a new model for [AI regulation](/category/policy) on July 14: a FINRA-style self-regulatory body with mandatory 30-day pre-release safety reviews for frontier models. The proposal lands in a policy vacuum where governments have legislated slowly and AI companies have governed themselves. Hassabis outlined the framework in an interview published across Axios, The Verge, and Techmeme. The central idea: an independent, industry-funded regulator modeled on FINRA — the Financial Industry Regulatory Authority that oversees US broker-dealers — would require AI labs to submit frontier-class models for safety assessment 30 days before public release. The rule would apply to any model meeting a capability threshold, "no matter their country of origin or whether they are open or closed." That's a significant caveat. It means DeepSeek, [Mistral](/glossary/mistral), Meta's open-weight releases — every major model, regardless of licensing — would fall under the same pre-release review. If risks are deemed sufficiently high, the regulator could recommend a government-mandated development slowdown. Not a voluntary pause. A mandated one. ## Why FINRA as the Model FINRA is a self-regulatory organization — technically private, but operating under SEC oversight with statutory authority to write and enforce rules. It's funded by the industry it regulates. It conducts examinations, imposes fines, and can bar individuals from the securities industry. Hassabis's choice of FINRA rather than an FDA-style pre-market approval model is deliberate. FDA approval takes years and costs hundreds of millions. A FINRA model would be faster, cheaper, and more adaptive — but critics argue it would also be more captured by the industry it oversees. Ian Hogarth, a UK [AI safety](/glossary/ai-safety) researcher, praised Hassabis for engaging on hard regulatory questions. Others were less charitable, describing the proposal as "regulation dressed in techno-jargon that preserves incumbent advantage." The 30-day window is the operational core. It's long enough for structured safety testing — red-teaming, capability benchmarking, alignment checks — but short enough that it doesn't cripple the release cadence frontier labs have settled into. [OpenAI](/glossary/openai), [Anthropic](/glossary/anthropic), Google, and Meta each ship major models every 2 to 4 months. A 30-day hold would slow nobody down by more than one cycle. ## The Reaction The split is predictable. Frontier labs that already invest heavily in safety — Anthropic and Google DeepMind chief among them — support structured pre-release review because it raises costs for competitors who don't. Smaller labs and open-source advocates see it as a barrier to entry dressed as a safety measure. The open-source community flagged the "open or closed" language immediately. A 30-day pre-release review for open-weight models raises uncomfortable questions. Who does the review? What happens if a model is flagged as dangerous after release? Once weights are public, there's no recalling them. Meta, which released [Llama ](/compare/llama-4-vs-deepseek-r1)models openly, hasn't commented. Neither has Mistral. DeepSeek, the Chinese lab that open-sourced competitive models and triggered a US policy scramble, is unlikely to submit to a US-led FINRA body voluntarily. Hassabis's proposal also sidesteps the enforcement problem. FINRA works because every US broker-dealer must register with it to operate legally. No equivalent requirement exists for AI model release in the US — and Congress has shown little appetite for creating one. ## FAQ #### Q: What's the actual proposal? A: A FINRA-style industry-funded regulator that reviews frontier AI models 30 days before release and can recommend government-mandated development slowdowns. #### Q: Would it apply to open-source models? A: Yes. Hassabis explicitly said the framework covers models "whether they are open or closed." This is one of the most contested parts of the proposal. #### Q: Who supports this? A: AI safety advocates and frontier labs that already conduct pre-release safety testing. It raises costs for competitors who currently skip those steps. #### Q: What's the biggest obstacle? A: Enforcement. FINRA has statutory authority over broker-dealers. No equivalent legal framework exists for AI model release in the US, and Congress hasn't shown interest in creating one. ## Related Articles Get AI news in your inbox Daily digest of what matters in AI. ## Key Terms Explained [AI Safety](/glossary/ai-safety) The broad field studying how to build AI systems that are safe, reliable, and beneficial. [Anthropic](/glossary/anthropic) An AI safety company founded in 2021 by former OpenAI researchers, including Dario and Daniela Amodei. [DeepMind](/glossary/deepmind) A leading AI research lab, now part of Google. [LLaMA](/glossary/llama) Meta's family of open-weight large language models.