Demis Hassabis, head of Google DeepMind, has suggested the establishment of a standards body in the U.S. to evaluate the most powerful AI models prior to their release. He estimates that artificial general intelligence (AGI) could emerge "within a few years," with an impact comparable to the discovery of electricity or fire.
— Demis Hassabis (@demishassabis) July 14, 2026
Hassabis defines AGI as a system possessing all the cognitive abilities of the human brain. He believes this technology cannot even be compared to the internet or mobile communication.
"When we look back on this time in the coming decades, I think we will realize that we stood at the foothills of the singularity — nothing less than the beginning of a new era for humanity," Hassabis wrote.
The DeepMind chief believes AGI could accelerate the development of medicines, new sources of clean energy, and advanced materials. However, he noted that AI capabilities are evolving faster than our understanding of the associated risks.
Current advanced models already pose threats in cybersecurity. As their capabilities grow, biological, nuclear, and other risks may increase. Hassabis also warned of the emergence of more autonomous systems capable of executing complex sequences of actions and recursively improving themselves. Additional technical safeguards will be necessary to maintain control over these systems.
New Body to Define Advanced Models
Hassabis proposed creating a Frontier AI Standards Body modeled after FINRA. This new entity could operate as a public-private partnership or a self-regulatory organization under federal oversight, primarily funded by the AI industry.
The leadership should include independent technical experts and representatives from the open-source community. The body will develop evaluation protocols in collaboration with federal agencies and national laboratories in the U.S.
Special tests will determine whether a model qualifies as advanced. These assessments will cover cybersecurity, biological threats, and other high-risk areas. Separate evaluations will identify attempts by models to circumvent protective measures, mislead users, or conceal their actions.
Mandatory Evaluation Possible
Initially, companies will voluntarily submit advanced models for evaluation no later than 30 days before their release.
Once the system's effectiveness is confirmed, the process could be codified into law. Advanced models would then be required to pass evaluation before entering the U.S. market. Hassabis also suggested creating a market for independent auditors and using the U.S. system as a foundation for future international standards.
In June, U.S. President Donald Trump signed an executive order for the voluntary evaluation of the cybersecurity capabilities of advanced AI models. Companies will be able to grant authorities access to these models for up to 30 days before sharing them with external trusted partners. The document explicitly prohibits interpreting this mechanism as mandatory licensing or obtaining prior approval for model releases.
In his essay, Hassabis acknowledged that the further development of technology remains uncertain and that even experts disagree on various aspects.
Earlier, reports indicated that the Trump administration asked OpenAI not to release GPT-5.6 to the public immediately due to safety concerns. Sam Altman's company initially provided the model to a limited number of clients.
In July, U.N. Secretary-General António Guterres called for universal rules for advanced AI systems, to protect children from chatbot risks, ban lethal autonomous weapon systems, and expand access to computing resources for developing countries.
In July, participants in a march in San Francisco demanded that OpenAI, Anthropic, and Google DeepMind coordinate to pause the training of more powerful models, suggesting that the freed resources be redirected towards safety.
That same month, former OpenAI researcher Daniel Kokotailo proposed slowing the AI race until 2040.