Google DeepMind’s chief executive, Demis Hassabis, used a LinkedIn post on Tuesday to make a bold pitch: the AI industry needs a dedicated standards organization to assess frontier‑model releases. He likened the proposed entity to the Financial Industry Regulatory Authority, suggesting it could test cutting‑edge models, develop best‑practice guidelines, and ultimately become a gatekeeper for U.S. deployment.

Hassabis’s plan calls for labs to voluntarily submit their models to the standards body up to 30 days before a public launch. If the review protocol proves effective, he said formal requirements could follow, making the certification a prerequisite for market entry. The body would also coordinate with labs on post‑release vulnerabilities, ensuring rapid remediation when risks surface.

The idea emerges from criticism of recent government reviews of Anthropic’s Mythos and OpenAI’s Sol. Observers complained that those reviews lacked technical depth and operated behind opaque decision‑making walls. By handing assessment duties to an industry‑funded but independently run organization, Hassabis hopes to inject technical expertise while preserving innovation.

Funding would come from AI companies themselves, mirroring FINRA’s self‑regulatory model. Staffing would blend open‑source advocates, technical experts, and representatives from AI safety groups that could be outsourced for specialized risk evaluations. Hassabis argues that such a structure would keep pace with the rapid acceleration of AI capabilities and adapt to emerging threats.

The proposal also addresses political skepticism. White House AI adviser Sriram Krishnan, speaking for a16z, has dismissed the notion of an “FDA for AI,” arguing that a regulatory agency within the executive branch would be infeasible. Hassabis’s self‑regulatory approach sidesteps that concern by positioning the standards body as an industry‑backed, government‑supported entity rather than a federal agency.

If adopted, the framework could reshape how frontier AI models move from the lab to consumers. Labs would gain a clear pathway to compliance, while regulators would acquire a technically proficient partner to vet high‑risk systems. The balance between oversight and innovation, a point Hassabis stressed, would hinge on the body’s ability to evolve alongside the field.

Questo articolo è stato scritto con l'assistenza dell'IA.
News Factory APP - notizie agentiche per potenziare il tuo SEO e AEO.