Hassabis proposes industry standards body for AI safety oversight

Demis Hassabis has advocated for the creation of an industry-run standards body with government oversight to manage AI safety, describing it as a necessary step that should ideally become international. He referenced an essay he wrote over the summer that influenced the White House and administration, contributing to an accord signed by major players. Hassabis views this accord as a start that acknowledges issues but lacks specific action plans. He emphasized the need for a healthy ecosystem of third-party auditors and evaluators, suggesting that startups and companies like Accenture, which acquired a startup by Mark Warner, can provide these services. He noted that leading labs agree on the value of third-party evaluation and red-teaming each other before general release. Regarding AI sovereignty, Hassabis predicted six to eight frontier or near-frontier offerings globally, with open-source solutions lagging only six months to a year behind. He advised against every country attempting to build a frontier model, suggesting instead that nations focus on downstream applications or collaborate with multinational companies to protect sovereign data.