US could set up "Standards Body" to regulate frontier AI models, says Google DeepMind CEO
New Delhi, July 15
The United States could establish a new Standards Body modelled on a federally overseen public-private partnership or self-regulatory organisation with a board that includes independent leading technical experts and open-source representatives.
According to Demis Hassabis, Co-Founder & CEO, Google DeepMind, the US is well positioned to take the first step in developing such a framework given its economic and technical standing.
"The rapid progress we're seeing in AI requires a new approach to testing frontier AI model capabilities...The US is well positioned, given its economic and technical standing, to take the first step in developing such a framework. It could establish a new Standards Body modelled on a federally overseen public-private partnership or self-regulatory organisation...," Hassabis said on X.
The proposal aims to create a dynamic, adaptable, and rigorous framework to test frontier AI model capabilities amid rapid technological progress.
Hassabis mentioned that the proposed entity, similar to the Financial Industry Regulatory Authority (FINRA), would require substantial funding coming mostly from the industry to attract world-class technical talent and provide necessary compute resources for large-scale testing.
The Standards Body would develop assessment protocols and collaborate with federal agencies and the US National Labs to conduct testing in areas relevant to national security. A model would qualify as "Frontier-class" by meeting specific benchmarks determined by the body, which would be regularly updated to match evolving capabilities.
Hassabis explained that organisations with these models would be deemed "Frontier Labs" and encouraged to adopt best practices, including publishing model cards, maintaining strong internal cybersecurity, vetting personnel, and resourcing safety research.
Initially, Frontier Labs would voluntarily share models with the Standards Body for review up to 30 days before release. Once the assessment protocol proves effective and robust, formalisation could follow, requiring Frontier Models to pass the review to be deployed in the US market. Labs would also work with the body to address critical post-release vulnerabilities.
"The strength of this approach is it would be technically focused, while at the same time supporting innovation and incentivising responsible behaviour," Hassabis said.
"It is designed to keep up with the field's acceleration and adapt to the biggest risks as they are identified, and could be ratcheted up if the seriousness of the situation demands, including coordinating a slowdown in development among the Frontier Labs if deemed necessary," he added.
Model assessments would include rigorous scientific evaluations of capabilities in cybersecurity, biological threats, and high-risk domains. Specific agentic AI tests would look for attempts to bypass safety guardrails or signs of deception, ensuring practices like digital watermarking and generating human-readable output tokens.
Hassabis noted that the evaluations would be updated quarterly at the start, with outdated benchmarks deprecated. The body would eventually build technical capacity to create independent held-out tests to prevent overfitting, promoting an ecosystem of third-party auditors with the US government.
"AGI has the potential to be the ultimate tool for advancing science and medicine, and to drive enormous productivity gains and economic growth," Hassabis stated.
"But in order to achieve this, we need to get the technical foundations right by coordinating around a shared global framework, using the most rigorous scientific methods, and bringing the best minds together to work on the challenges we face," he noted.
The framework would apply to Frontier-class models regardless of their country of origin or open-source status, while non-frontier models from startups or academia would remain exempt. The initiative seeks to provide a starting point for international standards to manage serious risks while ensuring global access to AI benefits.
— ANI
Reader Comments
This is exactly what we need. With AI advancing so fast, self-regulation by tech companies isn't enough. A standards body with real technical expertise and independence could prevent the kind of catastrophic failures we've seen in social media. The 30-day review window might be tight, but it's a start.
Does anyone else find it ironic that Google DeepMind, which is literally at the forefront of AI development, is proposing regulation? It's like the fox designing the henhouse. Still, I appreciate Hassabis acknowledging the risks—especially with agentic AI and cybersecurity threats. But let's be real: this framework will mostly benefit Big Tech while startups in countries like India will struggle to comply.
Finally, some serious thinking about AI governance! The FINRA comparison makes sense—having an industry-funded but federally overseen body could strike the right balance. The quarterly updates to benchmarks show they understand how fast this field moves. My concern is enforcement: what happens when a Frontier Lab refuses to comply? We need teeth, not just suggestions.
As someone working in AI research in Bangalore, I see both sides. On one hand, we desperately need standards to prevent misuse—especially with deepfakes and misinformation which is already a huge problem here. On the other hand, this feels very US-centric. What about data sovereignty? India's DPDP Act already has strict data localization requirements. If these standards conflict with our laws, will our AI companies be locked out of the US market? 🤔
The part about coordinating "a slowdown in development" is unprecedented and frankly a bit scary. How would that even work globally? Would China or Russia agree
We welcome thoughtful discussions from our readers. Please keep comments respectful and on-topic.