DeepMind CEO Urges US-Led Testing of AI Models

Demis Hassabis, Co-Founder and CEO of Google DeepMind. Credit: Demis Hassabis/LinkedIn

Google DeepMind CEO Demis Hassabis highlights the dangers of artificial general intelligence and advocates for a US-led governance framework

Google DeepMind’s CEO, Demis Hassabis, has proposed a US-led regulatory body to evaluate frontier-class AI models. He emphasized the urgent need to mitigate risks, including those related to nuclear and biological threats, as humanity approaches artificial general intelligence (AGI).

Demis Hassabis is a Nobel laureate and co-founder of Google DeepMind. He shared the Nobel Prize in Chemistry with colleague John Jumper, who recently announced his move to Anthropic. Demis holds a PhD in Cognitive Neuroscience from University College London.

While issuing serious warnings about AI threats, he expressed optimism about its potential: “We could even reach a point where resources are no longer the limiting factor for human progress, leading to an amazing new era of abundance,” he stated, arguing that AI will help solve some of humanity’s most pressing challenges.

Key concerns regarding AGI

Demis noted that leading AI firms are currently engaged in an intense, multi-layered commercial and geopolitical race. This competition involves his own Google DeepMind alongside rivals such as OpenAI, Anthropic, and DeepSeek.

“We’ve already seen the challenges frontier models pose for cybersecurity, and other threats including nuclear and bio risks may soon emerge as capabilities continue to advance,” Demis wrote on his Substack blog.

Demis added: “We must use this precious window before AGI arrives to shape this technology for the benefit of all humanity.”

He also highlighted potential risks from areas that are currently unknown: “On the horizon, we will need robust safeguards to maintain control of increasingly agentic, recursively self-improving systems - and tackle unknown issues that will only become clearer over time.”

This warning follows a recent US government directive, citing national security concerns, to suspend all foreign access to Anthropic's Fable 5 and Mythos 5 models, a restriction that has reportedly since been lifted.

An artist’s illustration of AI. This image explores how AI can be used to progress the field of Quantum Computing. Credit: Google DeepMind/Unsplash

Demis’ views on AGI are supported by other experts and think-tanks. The Center for International Relations and Sustainable Development warned in an article titled Why AGI Should be the World’s Top Priority that without national and international regulation, humanity risks losing control of a non-biological intelligence beyond human understanding, awareness, and control.

Notably, the Bulletin of the Atomic Scientists, which in January 2026 set the Doomsday Clock to 85 seconds to midnight—the closest it has ever been in its history—also called for urgent action to establish international guidelines on AI use.

Demis Hassabis, Google DeepMind CEO and Co-Founder (left) with John Jumper, former Vice President of Google DeepMind (right) | Credit: Jon Kopaloff / Getty Images

“We must use this precious window before AGI arrives to shape this technology for the benefit of all humanity. ”Demis HassabisCo-Founder & CEO, Google DeepMind

Proposal for US-led regulation

Demis argued that the US, given its economic and technical standing, is well-positioned to take the first step in developing a framework that employs a dynamic, adaptable, and rigorous approach to testing frontier models.

Demis’ proposal is detailed and specific. “It could establish a new Standards Body modelled on a federally overseen public-private partnership or self-regulatory organisation, much like the Financial Industry Regulatory Authority.

“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.”

“Initially, Frontier Labs would voluntarily share models with the Standards Body for review up to 30 days before release. Once the assessment protocol is shown to be effective and robust, formalisation could quickly follow, meaning that Frontier Models would be required to pass it to be deployed in the US market.”

Demis added that this US-initiated effort would provide a strong foundation for creating shared international standards on frontier AI.

Nik Kairinos, CEO & Co-founder, RAIDS AI, commented on the development: “But a global AI watchdog cannot be led by the agenda of any one nation.

Nik Kairinos, CEO & Co-founder, RAIDS AI. Credit: Nik Kairinos/LinkedIn

“AI does not respect borders, and nor should the rules designed to govern it. A multinational approach is essential to ensure any AI regulations achieve the necessary trust, cooperation and buy-in to work internationally."

Recent risks from frontier models

Demis’ warning follows Anthropic’s publication of a report, Agentic Misalignment in Summer 2026, detailing observations of alignment model failures in high-stakes simulations. The report utilized frontier models from Google DeepMind, OpenAI, Chinese AI company DeepSeek, and Anthropic’s own models.

The misalignment issues listed include models pursuing their own motivations beyond user instructions and what Anthropic terms ‘harmful compliance,’ where the user's request is followed but results in harm.

Issues cited by Anthropic include covert sabotage, an agentic misalignment failure where models secretly interfere with code to undermine user intent. Anthropic also found that some models would assist users in activities resembling white-collar crime.

Other risks from recent frontier models cited include coaching human proxies to whistleblow, where models leak confidential safety information externally, or steer humans toward doing so, as well as motivated mislabelling.

Anthropic stated in its report on the Alignment Science Blog that the purpose of these case studies is to help developers and evaluators measure similar failures and build targeted safeguards.