Frontier AI poses unprecedented risks and urgently requires international regulation and public debate
Source: https://www.theguardian.com/profile/richard-hermer. "When AI is subject to less regulation than your fridge, something is startlingly wrong | Richard Hermer | The Guardian." October 12, 2026. www.theguardian.com
The Gist
The author, a former UK attorney general, argues that advanced AI could threaten humanity's survival and our ability to control our own lives, yet it is barely regulated. Companies and rival countries won't police themselves, so governments need to act together through international law and the public needs to demand a say, before it's too late.
Conclusion
Frontier AI must be rigorously regulated now, through international law and cooperation and an urgent, inclusive public debate, because it poses an existential risk and challenge to humanity and the window for effective action is limited.
Premises
- Frontier AI poses two kinds of risk: an existential risk (enabling mass-casualty harms such as biological weapons, or AI itself deciding to dispense with humans) and an existential challenge (deprivation of billions of people's meaningful agency over their lives, with profound effects on economies, security and happiness).
- AI is developing to a level humans do not understand and may eventually be unable to control, and its complexity and potential devastation exceed those of past failures such as the 2008 financial crash.
- We do not know the precise degree of risk, but even a significant chance of catastrophic harm is unacceptable, since we would not allow a toaster with a 10% chance of electrocuting users to be sold.
- Frontier AI labs are currently less regulated than household appliances, and self-regulation cannot be relied on because companies, markets and the US-China geopolitical race incentivise developing new iterations as fast as possible; voluntary codes provide little reassurance.
- Because AI development is global and driven by state rivalry, national action alone is insufficient, and effective governance requires international law and cooperation among the US, China, like-minded state coalitions and the tech sector.
- Control over AI is concentrating in two states and a small number of tech-company leaders, so democratic legitimacy requires an urgent public debate to determine who controls AI and whom it serves.
- There is a limited time window to act, and recent events (the Hugging Face hack, a resignation at Anthropic) have created a rare opening in public discourse.
Assumptions
- The risks of frontier AI (including extinction-level outcomes) are real and plausible enough to justify precautionary regulation despite acknowledged uncertainty.
- Binding regulation and international law can be designed, agreed and enforced effectively for fast-moving AI technology, including with rival powers.
- Slowing or constraining frontier AI development will reduce risk without causing offsetting harms, such as ceding advantage to less scrupulous actors.
- Product-safety analogies (toasters, fridges) transfer meaningfully to general-purpose AI systems.
- Public debate and democratic engagement can influence outcomes in the face of powerful state and corporate interests.
- Expert warnings from AI insiders and figures like Bill Gates are credible indicators of genuine risk.