19. September 2026 Geneva

Please find below the Op-ed in the Neue Zürcher Zeitung (NZZ) on 14 September 2026 by Daniel Stauffacher of ICT4Peace:

When It Comes to the Extreme Risks of Artificial Intelligence, Switzerland’s Hour Has Come

On the same day, 29 states in Shanghai and the Vatican demanded that artificial intelligence must never threaten peace and security. What is missing is a neutral place of convergence – Switzerland should provide it.

Guest commentary published in the Neue Zürcher Zeitung (NZZ) on 14 September 2026. English translation.

16 July 2026 may go down in the history of technology policy. In Shanghai, 29 states founded the World Artificial Intelligence Cooperation Organization (WAICO) to promote global AI governance. On the same day, the Vatican gathered two hundred experts, among them several Nobel laureates; their declaration calls for effective human control over nuclear decisions and for nuclear disarmament. Two different senders, one shared conclusion: AI must never be used in any way that threatens peace and international security.

National laws on parallel tracks

Behind these appeals lies a broader trend. On 2 June, US President Trump issued an executive order giving federal agencies 30 days to voluntarily review new AI models for cybersecurity risks before release.

China, for its part, published on 8 May 2026 the first standalone set of rules treating AI agents as a distinct class of systems. And it is not only the two leading AI powers that are legislating: Brazil, the EU and South Korea have laws in force; India is working on a parliamentary act against systemic AI risks. However different their capabilities, they are moving in parallel towards AI safety.

The past months have shown that this is not science fiction. First: offensive cyber capabilities. Claude Mythos, Anthropic’s frontier model, found zero-day vulnerabilities in major operating systems without human guidance and broke out of its test environment; after its release in June, the company had to withdraw it at Washington’s demand.

Second: the easier development of biological and chemical weapons. Early this year, a British organisation demonstrated that Grok, the AI chatbot developed by xAI, explained how to produce ricin, chlorine gas and nitrogen mustard. Add to this the nuclear dimension: as early as November 2025, the First Committee of the UN General Assembly adopted a resolution on AI risks in nuclear command, control and communication systems.

Third: the irreversible loss of human control. In mid-July, an agentic AI system from OpenAI broke out of its test environment during a hacking test and penetrated the infrastructure of the Hugging Face platform. OpenAI noticed the attack only after the targets had discovered it themselves.

A minimum of rules, a maximum of effect

What follows from this? Certainly not a call for centralised world regulation – geopolitically it is neither necessary nor possible. The report “The Essential Convergence” by the Strategic Foresight Group, presented in Geneva on 6 July, points to a different path: a minimum of common rules for those extreme risks that are global, unpredictable, potentially irreversible and catastrophic. National laws already show how to prevent them: limits on computing power, mandatory evaluations before new models are released, reporting of AI incidents that warrant investigation. Convergence on such minimum standards is feasible because national legislation is moving in that direction anyway.

What is missing is a neutral body that brings everyone to the table. No country is better equipped for this than Switzerland. Geneva is the world centre of humanitarian governance – and the extreme risks of AI have profoundly humanitarian consequences. The Geneva Conventions were born on Swiss soil; during the Cold War, Geneva provided the neutral setting for arms-control talks such as the SALT negotiations. In this tradition, Switzerland, as host of the AI summit, can bring states together for a global understanding on AI safety and proclaim that convergence in Geneva.

Rarely has there been such broad consensus both on the risks and on the measures required. Until now, Switzerland has mostly helped to solve problems after disaster had struck. This time it can prevent a catastrophe before people come to harm. It should not let this window of opportunity pass unused. For this to become binding convergence, a neutral venue is needed – Switzerland, host of the next AI summit in Geneva in June 2027.

Daniel Stauffacher is the founder of the ICT4Peace Foundation and a former Swiss Ambassador.

Here the Original German Version: Stunde_der_Schweiz_KI

English Version Stunde_der_Schweiz_KI_EN

French Version Stunde_der_Schweiz_KI_FR

Italian Version Stunde_der_Schweiz_KI_IT