Google DeepMind's CEO Demis Hassabis has proposed the establishment of a frontier AI standards body modeled on FINRA, to be led by the US and operating as a public-private partnership under federal oversight.
Hassabis argues that AGI may arrive within years, and that frontier models already pose cybersecurity risks that outpace the industry's and society's capacity to manage them; the standards body would evaluate and certify frontier-level AI systems before their release, and could coordinate development slowdowns between frontier labs if circumstances warrant.
The proposal comes amid broader calls from economists and AI researchers for urgent policy action to manage AI's transformative economic impact.
What happened
Google DeepMindのデミス・ハサビスCEOが7月14日、フロンティアAI標準化機関の設立を提案するエッセイを公開しました。FINRAをモデルに、連邦政府の監督下での官民パートナーシップまたは自主規制組織として設立し、独立した技術専門家やオープンソース界の代表を理事会に含める案です。
Why it matters
ハサビス氏はAGI（汎用人工知能）が「おそらくあと数年で実現する」と指摘し、フロンティアモデルがもたらすサイバーセキュリティの課題は既に顕在化していると警告しています。業界の進歩が技術への理解を上回るペースで進んでいるため、リスク軽減に必要な時間と余地を確保できていないと述べており、この機関はその解決策と位置づけられています。
What to watch
同機関はベンチマーク基準を満たしたモデルを「フロンティア級」、それを持つ組織を「フロンティアラボ」と認定する予定です。当初、フロンティアラボはリリースの最大30日前に自主的にモデルを提出してレビューを受け、有効性が実証されれば審査通過を米国市場での展開要件とすることも視野に入れています。
Ask the AI about this article →
Hassabis's proposal reflects growing concern within the AI industry about the pace of development outstripping regulatory and risk-management capacity. The essay frames AGI as arriving imminently—within years—and compares its potential impact to a tenfold amplification of the Industrial Revolution in scale and speed. The CEO explicitly warns that frontier models already pose cybersecurity challenges, citing unspecified examples, and that nuclear and biological risks may soon emerge. By anchoring his proposal in an existing financial-sector regulatory model (FINRA), Hassabis suggests a pragmatic template rather than an untested framework.
The timing of the proposal is noteworthy: it follows Hassabis and Anthropic CEO Dario Amodei's participation in a G7 meeting in June (per CNBC), where they called for US-led coordination on AI rules and standards. Amodei has separately advocated for binding pre-release review in his own blog. The broader context includes the US government's June export controls on Anthropic's models (since lifted as of June 30), which Hassabis references obliquely when noting that frontier models' cybersecurity impacts are "already being seen." One day after Hassabis published his essay, over 200 economists and AI researchers, including 16 Nobel laureates, released a statement warning that AI could trigger economic transformation exceeding the Industrial Revolution in scale but arriving in far less time, calling for urgent policy and institutional action—though Hassabis did not sign that statement.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
