DeepMind bigbrain calls for America to set AI standards before it's too late

ai and ml

Industry, regulate thyself

Google DeepMind boss Demis Hassabis is calling for the US to establish a robust frontier AI model review process because, according to him, artificial general intelligence (AGI) “is probably only a few short years away" and we've got to figure it out before it's too late.  

That “probably” is doing a lot of heavy lifting in Hassabis’ lengthy, early-morning Tuesday post on X. Like commercially viable fusion power and practical fault-tolerant quantum computing, AGI seems perpetually asymptotic to the present moment, with needed technological advancements always the next step in the process. 

No amount of hyperbole - not even a Nobel laureate with skin in the game predicting an impact “perhaps 10x of the Industrial Revolution at 10x the speed” - makes the AGI timetable any more certain. Recall that Hassabis predicted in the beginning of 2025 that human trials of AI-designed drugs would come that year, and that still hasn’t happened either. 

Regardless of the questionable prognostications, the main argument in Hassabis’ essay – that we need to establish international standards for classifying AI safety and risk – is worth serious discussion, and his arguments for it are sensible. 

“I’m confident that mitigating the technical risks related to AI is a challenge we can collectively address, but only if we give ourselves the time and space to get this next crucial step right,” Hassabis said. “Currently, as a field and as a wider society, we aren’t doing that.”

He’s got that right, at least. 

Hassabis proposes that the US ought to create a new standards body to evaluate frontier AI models in the same vein as the Financial Industry Regulatory Authority (FINRA), the private, industry-funded self-regulatory organization that oversees US broker-dealers under SEC supervision and is charged with protecting investors and safeguarding market integrity.

The DeepMind CEO’s vision for an AI industry regulatory authority would begin with a board of tech experts and open-source representatives funded by the AI industry itself – a move that would “attract world-class technical talent” and ensure the body had enough hardware to run its various frontier model tests. FINRA is funded in a similar manner by the industry it purports to regulate, which has opened the door to criticism that it's basically a toothless insiders' club.

Funding and conflict-of-interest concerns aside, the main task of the AI standards body Hassabis proposed would be to establish an assessment protocol for frontier AI models, and to determine what qualifies as a frontier model based on benchmarks it would also be responsible for setting. 

“Organisations with ‘Frontier Models’ as defined by those benchmarks would be deemed ‘Frontier Labs’, and be encouraged to adopt best practices, such as publishing model cards with technical details, maintaining strong internal cybersecurity, vetting key personnel, and providing sufficient resourcing for safety and security research, and more,” Hassabis explained.

Hassabis proposed that submitting to such evaluations would be voluntary for frontier labs, at least at first – the model he proposes would give labs the option to voluntarily submit models for review 30 days before release until whatever assessment protocol the group makes up proves to be effective. He further proposes that frontier labs should be allowed to help the standards body develop those benchmarks – again, at least for a while until it gets up to speed. “Eventually the Standards Body should build up the technical capacity to create its own held-out tests independent of the Labs to prevent overfitting.” 

The hope is that this US-led effort would eventually lead to international frontier AI assessment standards, though whether the international community would stomach another imposition of US rules on the global stage is questionable at this point. 

The Trump administration has already proposed something similar for testing frontier AI models before their broader release. President Trump signed an executive order in early June directing the National Institute of Standards and Technology, alongside several other federal agencies, to develop a voluntary framework that would allow the government to review covered frontier models for up to 30 days before they are shared with select trusted partners. The effort is primarily aimed at evaluating advanced cybersecurity capabilities, and the order also calls for the development of classified benchmarks for assessing frontier AI models.

Some critics argued the framework could give the government undue influence over which companies receive early access to frontier models, pointing to provisions governing collaboration with "select trusted partners." An independent standards body could avoid some of those concerns.

Given concerns that financial industry pressure affects the outcome of FINRA work, however, even an independent AI standards body may not be enough to prevent financial influence from skewing its decisions. 

“The future is not yet written,” Hassabis concluded in his note. “We must use this precious window before AGI arrives to shape this technology for the benefit of all humanity.”

Whether or not AGI is actually coming, the AI industry definitely needs someone to kick it into shape. The question of who can do it fairly – the industry itself or the government – isn’t one that’s been adequately answered yet. ®