Deutschlandfunk.de Interview with Daniel Stauffacher of ICT4Peace on Regulating and Pacing of AI Frontier Models
Interview by Manfred Götze of Deutschlandfunk.de with Daniel Stauffacher
Following the Op-ed in the Neue Zürcher Zeitung (NZZ) by Daniel Stauffacher of ICT4Peace.org, Manfred Götzke of the Deutschlandfunk.de made this Radio Interview with him on Regulation and Pacing of AI Frontier Models and AI Risks and Switzerland’s possible role in hosting Global Negotiations in Geneva.
Kindly find the English Translation of the Interview here and here the French and Italian translations.
Interview with Daniel Stauffacher, ICT4Peace Foundation
Broadcaster: Deutschlandfunk (German public radio) Programme: Europa heute
Presenter: Manfred Götzke, Deutschlandradio, Berlin studio Aired: Friday, 18 September 2026, approx. 09:12 CEST Duration: 10:06 min. Original language: German. English translation from the German broadcast transcript by ICT4Peace with the help of Claude.
Manfred Götzke: Ten to twenty percent – that is how high the heads of the major AI companies rate the risk that AI could wipe us out in the coming years. The CEOs of OpenAI and Anthropic, even Elon Musk, are therefore calling for regulation and a slowdown of development. But that is unlikely to achieve much unless all the relevant companies and states go along with it. What would be needed is an international agreement – with Switzerland as a neutral body that brings everyone to the table. That, at any rate, is how Daniel Stauffacher sees it. He is the founder of the ICT4Peace Foundation, an NGO that works for a peaceful digital world. Before the programme I asked Daniel Stauffacher how surprised he was by the fairly unanimous demand from the AI companies that AI development must be slowed down.
Stauffacher: This is a very particular issue, because at the moment there is a real muddle among these large companies. Some want a slowdown, others do not. My view is that a slowdown is ultimately not possible at the moment – for geopolitical reasons, given the competition between the United States and China, and generally because of the investment dynamics. So I treat this with caution.
Manfred Götzke: Do you see an international consensus that AI should or must be regulated, or at least that its development must be slowed, so that AI does not wipe us out in the coming years, by the end of the decade – as one could often read this week?
Stauffacher: One has to distinguish. On regulation, I see a growing consensus, a convergence. You can see it not only in declarations but also in actions: we have the EU AI Act, we have the Council of Europe convention, we have national laws. Then we have the Chinese initiative for the World Artificial Intelligence Cooperation Organization, and at UN level the Global Dialogue on AI Governance and the Scientific Panel. There I see convergence. On slowing down, as I said, I am not sure. But where I also see a growing consensus is on the extreme risks of frontier models: misuse for biological and cyber weapons, loss of control, military autonomy. There, even the competitors actually have an interest in seeing this contained in some way. And it is precisely with these frontier models that there have been very problematic, worrying incidents.
Manfred Götzke: You say that in an international regulation of AI, Switzerland could play a relevant, an important role. In what way? What do you have in mind?
Stauffacher: It is partly a happy coincidence: next year Switzerland is organising a global AI summit, following on from the summits that began in Bletchley and continued in Seoul, Paris and New Delhi. There is not much time left in this process, but we still have the chance to discuss this question of convergence and possible regulation. Switzerland has a certain tradition as a bridge-builder in UN processes, as host country of various UN organisations. And Switzerland could perhaps help, against a possible bloc formation among the great powers on AI, to bring people and members together. But we can go into more detail if you like.
Manfred Götzke: Yes, gladly. What could regulation look like, and how could all the relevant players be brought to the table and to a consensus?
Stauffacher: I see it on three levels. The first level is the area of frontier models we have just discussed, with the greatest risks, which are really not in the interest of the great powers. There one could, for example, agree binding safety rules for the few developers of the most powerful models: testing before release, mandatory incident reporting, compute thresholds – and the rules follow the model’s capabilities, not its country of origin. The second level would be access and diffusion; this is something very important to me: a commitment to openness, to building compute capacity and expertise for countries without models of their own, so that the agreement is not perceived as a cartel of the big players. The third level would remain the national level – everything below that: the EU AI Act, Swiss legislation, minimum standards and so on. What one could perhaps also agree on are institutions: a scientific observatory, as we have in the climate debate, a joint testing and incident-reporting body, perhaps in future something like an IAEA. Realistically, Geneva will certainly not yet see detailed negotiations, but perhaps a political declaration, which would then be followed by a framework convention with protocols, along the lines of the Rio climate process.
Manfred Götzke: You have already hinted at it: Switzerland is, or would be, particularly well placed here, because other international agreements – on nuclear weapons, chemical weapons – were also negotiated and concluded in Switzerland.
Stauffacher: Exactly, that is the know-how Switzerland has. I also think it is important that Switzerland is neutral and has no frontier labs of its own, but has instead created the open-source model Apertus. That may also be an important element for bringing about, as an honest broker, neutrally, a consensus between all the blocs and the developing countries.
Manfred Götzke: Many consider AI to be just as dangerous as nuclear or chemical weapons, for which there are agreements. Could those treaties serve as a blueprint?
Stauffacher: Absolutely. Or, for example, the Basel rules for the large banks. That is also a model I have in mind, where one concentrates on the highest risks – the largest, systemically relevant banks – and regulates them tightly, while leaving the smaller banks somewhat freer. With that kind of gradation, starting from the greatest risks, where – as with the nuclear agreements – all sides have an interest, in deterrence and so on, one can find a consensus.
Manfred Götzke: Some heads of the big AI companies have argued for a slowdown and more or less asked for regulation. Trump does not want it. With China one does not quite know. How likely do you think it is that a viable agreement could come about in the coming months, perhaps next year?
Stauffacher: On regulation I have to say: the USA has in fact also regulated – I wrote that in my essay too – even though Trump goes back and forth. I think he meant the slowdown rather than a minimum of regulation. The Chinese are in fact also in favour of regulation. There are many examples where the USA and the great powers are in practice already regulating. But at the middle level it will be difficult to reach a consensus. On the major risks we discussed, however, where there is a danger for all participants, we should be able to reach a consensus.
Manfred Götzke: In June 2027, as you said, an international AI summit is taking place in Switzerland, in Geneva. And there, you say, the corresponding negotiations could begin.
Stauffacher: Yes. We will certainly not have an agreement in time for this summit; the time is too short. But we could possibly obtain a mandate from the world community, which says: yes, this is an important issue, and we task the UN or certain countries with taking up negotiations – first a framework convention, then protocols – and with launching a negotiation process. To already have something fully negotiated by next June seems unlikely to me. But a mandate should be possible, especially if the host country, Switzerland – my government – pushes hard for it.
Manfred Götzke: Switzerland as the country for international AI regulation – I discussed that with Daniel Stauffacher, former diplomat and founder of an NGO that works for a peaceful digital world.