antirez

ℏεsam

Aaron Levie

<h2>What Hassabis is proposing</h2>
<p>Google DeepMind CEO Demis Hassabis has published a proposal for an independent AI standards body modeled on FINRA, the industry-funded watchdog that oversees Wall Street. The core idea: frontier AI labs would submit models for review up to 30 days before release, with the body testing for dangerous capabilities including cyberattack potential, bioweapon risk, and deception.</p>
<p>Participation would start voluntary, then become mandatory once the process proves itself. The body would be funded by the AI industry but run independently, staffed by technical experts and open-source representatives. Most significantly, it would have the power to coordinate a slowdown across frontier labs if risks escalate enough to warrant it.</p>
<p>Hassabis says he’s already briefed the Trump administration, rival labs, and European officials. He cites a narrow window before AGI arrives and warns that bio and nuclear risks could emerge within 18 months.</p>
<h2>The broader convergence</h2>
<p>This isn’t just Hassabis. The heads of OpenAI, Anthropic, and Google DeepMind have each published memos calling for frontier AI regulation in the same period, which is a strange thing to see from companies that compete fiercely on everything else.</p>
<p>They agree on the basics: independent pre-release testing, a single standards body, US leadership on AI governance. Where they diverge is enforcement. Anthropic’s Dario Amodei wants an FAA-style agency with actual power to block releases. Altman is pushing for an IAEA-style international forum. Hassabis lands somewhere in between with the FINRA model.</p>
<p>The proposal drew endorsements from Microsoft, OpenAI, and xAI CEOs, plus praise from Satya Nadella and Elon Musk.</p>
<h2>The problems nobody has solved</h2>
<p>The framework has real gaps.</p>
<p><strong>China.</strong> There’s no mechanism to compel Chinese labs like DeepSeek to participate. The proposal leaves Washington’s export-control powers intact but offers nothing new on international enforcement.</p>
<p><strong>Regulatory capture.</strong> An industry-funded body overseeing the industry that funds it is a classic conflict of interest. The question of whether it can credibly act as an independent brake on the labs paying for it remains genuinely open.</p>
<p><strong>Open source.</strong> The proposal would apply to frontier-class models regardless of country of origin or whether they’re open or closed. With both Hassabis and Amodei pushing this direction, some form of restriction on open-weight models or fine-tuning and redistribution seems plausible. The proposal was partly inspired by the “Fable 5” incident, where US export controls abruptly disabled Anthropic models globally.</p>
<p><strong>Political environment.</strong> The Trump White House has already dismissed the idea of an FDA-style AI regulator. A FINRA model is a softer ask, but it’s still asking an administration that’s skeptical of AI regulation to endorse a new oversight body.</p>
<h2>The criticism from the open-source side</h2>
<p>Not everyone is impressed. The argument that a handful of large labs should control frontier AI, with open-weight models treated as a risk to be managed, strikes some as self-serving. The same arguments were made about open-source software, and they didn’t hold up.</p>
<p>Aaron Levie’s read is more charitable: a standards body is meaningfully different from a regulatory agency, and could move faster than government while still providing some coordination. His caveat is that even people inside the industry don’t agree on what the actual safety risks are, which makes building consensus around any framework harder than the joint memos make it look.</p>
<h2>What’s still unresolved</h2>
<p>Board composition, enforcement mechanisms, and China’s participation are all open questions. The proposal is a starting point, not a finished plan. Whether it becomes anything depends on whether the Trump administration engages with it at all, and whether the labs that endorsed it are willing to accept a body that might actually slow them down.</p>


Collections

Demis Hassabis of Google DeepMind has proposed a government-backed standards body for frontier AI models, covering both open and closed source regardless of country of origin, with provisions to slow AI development if risks exceed certain thresholds. Dario Amodei of Anthropic has expressed similar support, making two of the most prominent AI leaders aligned on the need for oversight. The proposal favors a standards body over a full regulatory agency to enable faster movement and closer industry collaboration. Critics like Aaron Levie warn that government-paced regulation could cost the US its competitive position. The deeper unresolved issue is that the industry itself lacks consensus on what safety risks actually exist, which threatens to make any resulting standards either ineffective or arbitrary.

Demis Hassabis calls for a frontier AI standards body, with mixed industry reaction