A call for a standards body to oversee the development of frontier artificial intelligence was voiced Tuesday by the CEO of Google DeepMind, an AI research lab.
"The Standards Body would be responsible for developing assessment protocols and working with appropriate federal agencies and the US National Labs to conduct testing in areas relevant to national security," Demis Hassabis wrote on LinkedIn.
He suggested the panel be modeled on a federally overseen public-private partnership or self-regulatory organization, like the Financial Industry Regulatory Authority (FINRA), and include independent leading technical experts and open-source representatives.
"At the moment, we are locked in an extremely intense, multilayered commercial and geopolitical race," he explained. "While these competitive dynamics fuel rapid progress and accelerate the incredible upsides, advances on the frontier are outpacing our understanding of the technology."
"Nobody in the world knows for sure what is going to happen from here, and even the experts disagree," he continued. "When there is a large degree of uncertainty and the stakes are this high, proceeding with cautious optimism is the sensible and correct strategy."
"That calls for public policy that promotes innovation while also incentivizing responsibility and security, fosters international collaboration on key safety issues, and encourages careful consideration of how AI is deployed for the benefit of society," he advised.
Case for an AI Standards Body
A standards body is certainly needed, declared Chris Canal, co-founder and CEO of EquiStamp, an AI safety company in Lewes, Del. "AI is a dual-use meta-technology," he told TechNewsWorld. "The same capabilities that accelerate drug discovery can lower the barrier to cyberattacks and bioweapons."
"Two things make pre-deployment standards non-negotiable," he said. "First, open-weight models cannot be recalled once released. The benefits and dangers an open-weight model provides exist permanently."
"Second," he continued, "labs like Anthropic and OpenAI already share non-public models with partners and testers before release, so a dangerous capability can be in circulation and cause harm before any government knows it exists."
Jeff Williams, CTO and co-founder of Contrast Security, a runtime security company in Los Altos, Calif., agreed that a standards body is needed to oversee AI development, particularly frontier AI. "These systems could ultimately pose risks comparable to nuclear, biological, or chemical weapons," he told TechNewsWorld. "We do not yet know how dangerous they will become, but the evidence is strong enough that waiting would be irresponsible."
He recommended that the approach be scientific, focused on measurable capabilities, and built on the work already underway in the security community. "The Open Worldwide Application Security Project [OWASP], in particular, has developed practical guidance across a broad range of generative and agentic AI risks," he said. "A standards body should fund, validate, and scale that work — not reinvent it behind closed doors."
Strengthen Existing Capacity
"We need rigorous, standardized testing of frontier models," added Michelle Lopes Maldonado, associate director of AI policy at the Information Technology & Innovation Foundation (ITIF), a research and public policy organization in Washington, D.C.
"The United States is already building that muscle," she told TechNewsWorld. "The Center for AI Standards and Innovation [CAISI] at NIST has pre-deployment evaluation agreements with most of the major labs."
"We need to strengthen and properly fund this existing capacity before standing up a new institution," she argued. "Ad hoc, lab-by-lab safety testing won't scale as capabilities advance."
That existing testing infrastructure raises a central question surrounding Hassabis' proposal: whether the United States needs a new AI standards body or should strengthen and expand the resources of institutions already evaluating frontier models.
Michael Bell, CEO of Suzu Labs, a provider of AI-powered cybersecurity services in Las Vegas, argued there is a need for accountability for specific, measurable harms. "We do not necessarily need a new bureaucracy standing between every model release and the market," he told TechNewsWorld.
"The security risks from frontier AI are real," he acknowledged. "I run a firm that red teams AI systems for enterprise and government clients, and the attack surfaces are expanding faster than most organizations understand. But there is a difference between targeted oversight of genuine national security risks and building a regulatory apparatus that slows down an entire industry based on capability thresholds defined by the same companies lobbying for the body."
"The question should not be 'do we need a standards body?'" he said. "It should be 'what specifically are we trying to prevent, and is a new institution the most effective way to prevent it?'"
A Better Template
Canal questioned modeling the new AI review framework on FINRA. "FINRA polices conduct after the fact in a domain where losses are financial and recoverable," he explained. "Finance is also the field that explicitly refuses to price catastrophe. Most insurance contracts carve out exactly the disaster categories AI could touch."
"The better template," he argued, "is something like the ASCE building code. When we approve a project that impacts society, like a bridge or building, we assess public benefit, set an explicit tolerated failure probability, and derive the requirements from it."
"That last part matters," he continued. "When the tolerance is explicit, the rule is the minimum necessary to meet it, not whatever a committee felt like."
"We should regulate frontier AI the way we permit nuclear plants, not the way we police stockbrokers," he asserted. "This holds whether the entity building the AI is a government or a private institution. Both have to abide by building codes, and codes are enforced and checked at multiple points in the building process — in blueprint/design, pre/mid/post construction, and then ongoing after project completion."
Hassabis' recommendation that the standards board be part of a public-private partnership is an important one. "A public-private partnership is essential because no single group has everything required," Williams explained.
"Frontier labs have the models, talent, and compute," he said. "Government brings authority, intelligence, and national-security resources. The independent security community brings adversarial expertise and openness."
"But this would not be a partnership among equals," he continued. "AI companies are investing billions and moving at extraordinary speed. Government moves slowly, while much of the security community still relies on volunteers. Unless independent research is funded at something closer to industry scale and speed, the frontier companies will define the agenda simply because everyone else is permanently trying to catch up."
Risks of AI Self-Regulation
Making the standards board a self-regulatory agency has advantages, too. "A self-regulatory organization can hire technical talent at market rates, move faster than statutory rulemaking and update benchmarks quarterly rather than once a decade," Maldonado explained. "FINRA works because industry insiders often understand the risks better than distant regulators do."
She conceded there are also disadvantages. For example, there's the threat of regulatory capture. "A body funded and informed by frontier labs may face constant pressure to set benchmarks that favor incumbents and raise rivals' costs," she said.
She added that FINRA may not be the best model for the standards board. "FINRA polices conduct in a mature industry with well-understood rules, while AI evaluation science is still young," she explained. "Independent audits already show inconsistent methodologies across assessors."
"Any board dominated by frontier companies would have a vested interest in building a body that protects them and gates new models from coming in," added Sean Campbell, co-founder of Cascade Insights, a provider of market research and strategic marketing services, in Portland, Ore.
"It's a classic fox-guarding-the-henhouse problem," he told TechNewsWorld.
He supported strong public-private partnerships, but added that strong global partnerships were needed, too. "Otherwise, it's like trying to control nuclear proliferation by making agreements with the 50 states," he said.
Standards Before AGI Arrives
Hassabis maintained that establishing AI standards now will smooth the way for the AI of the future, artificial general intelligence (AGI).
"There is both huge excitement and uncertainty around AI, and both are warranted," he wrote. "But the future is not yet written; we must use this precious window before AGI arrives to shape this technology for the benefit of all humanity."
"What we collectively do now will determine how the next phase of civilization unfolds," he continued. "By safely stewarding AGI into the world, we can enter a new golden age of scientific discovery and progress, and usher in a bright future of incredible human flourishing."
Loading ...
