ETH Zurich | Albert Einstein School of Public Policy Conference on Trustworthy AI for Human Rights, Democracy and Rule of Law.

Congratulations and thanks to ETH Zurich | Albert Einstein School of Public Policy for an inspiring Conference on Trustworthy AI for Human Rights, Democracy and Rule of Law.

In light of  some of the points raised during the Conference, ICT4Peace is pleased  to share two complementary papers on the governance of advanced and extreme AI risks and “Sovereign AI”. 

1. “The Essential Convergence” by the Strategic Foresight Group and launched in Geneva by ICT4Peace, Geneva Centre for Security Policy and International Federation of Red Cross and Red Crescent Societies – IFRC (https://lnkd.in/eEu3Q3qR)), argues that a narrow set of extreme AI risks (cyberattacks, biological and chemical weapons, mass manipulation, and loss of human control) are preventable but demand zero tolerance, and proposes a Global Compact built on five pillars: an international accord banning the most dangerous systems, a harmonised risk protocol, a global incident reporting exchange, a multilateral insurance facility, and a “two-key” authorisation system for releasing hazardous models
.
ICT4Peace with his partners is committed to advance the proposed Global Compact.

2. ICT4peace Foundation‘s own paper on Sovereignty, Access and Extreme Risks (https://lnkd.in/eWE6Y5fx) asks how such a Global Compact could be reconciled with the rise of “Sovereign AI.” It argues that the AI sovereignty trend, properly understood, helps separate what truly requires binding global agreement from what does not, and proposes a three-layer architecture modelled on the post-2008 financial reforms:

– A narrow binding core for extreme risks (incl. cyberattacks, biological and chemical weapons, mass manipulation, and loss of human control)
– A broader layer of voluntary technical cooperation: shared evaluation methods, common incident-reporting formats, and model documentation standards.
– Full national discretion on everything else: deployment rules, content governance, data policy, and industrial strategy.

Crucially, it holds that no binding regime on extreme risks will win broad support unless access is part of the same bargain, and it treats the AI divide facing the Global South as a core development question. Read together, the SFG report makes the case for what must be governed globally, while the ICT4Peace paper shows how that core can be made politically viable in a world of divergent national approaches.

ICT4Peace believes that these important topics should be discussed in international fora like the UN, and importantly also in the run-up to and during the Geneva AI Summit 2027. (https://lnkd.in/eAJsj4ZC)