Will there be waves on the shores of Lake Geneva? On June 17, in Evian, Emmanuel Macron brought together the bosses of AI companies, Google DeepMind, Anthropic and OpenAI, guests of honor at a tense G7. The impact of the Trump administration’s decision to cut off access to Anthropic’s latest models to companies and foreign nationals, a decision that led to the blocking of Mythos 5 and Fable 5, will certainly be on everyone’s minds. But “there is not much to expect from their discussions,” thinks Julien Nocetti, author of Les Gafam (Que sais-je?, 2026). The researcher associated with the Geopolitical Technology Center of the French Institute of International Relations returns for Libération to the consequences of this snub for Europeans.
Anthropic announced on June 12 that it had deactivated its latest Mythos 5 and Fable 5 models at the request of the Trump administration, which demanded that access be cut off for “any foreign national, inside or outside the United States”, including for its own “foreign employees”. How do you interpret this decision?
The affair can be read at different levels. The restrictions imposed on Anthropic’s two most advanced models mark a further step towards the generalization of the risk of kill switch [la capacité de désactiver l'accès à une technologie]. Today, access to an AI model can be cut off overnight. Tomorrow, it could be a cloud service, a cybersecurity tool, a software brick. Europe is discovering a new form of dependence: no longer just depending on components or servers, but on access authorizations. Next, the affair reveals the growing friction around visions of “digital sovereignty”. In fact, the approach embodied by the Trump administration is close to the Chinese statist vision. The latest American presidential decree, dated June 2, which invites American companies to give the government early access for thirty days to their latest AI models before public deployment shows this well. Finally, we must not lose sight of the framework of Sino-American technological competition, which guides a significant part of the Trump administration’s decisions in AI.
Beyond questions of sovereignty, is this sequence not the continuation of the battle between the Trump administration and Anthropic since February, when the company refused to allow its models to be used for domestic surveillance and lethal autonomous weapons?
This is an a priori attractive hypothesis: Anthropic, the white knight of AI made in America, would bow to the powerfully expressed will of the executive power… We cannot of course exclude the motivation of revenge – it is a psychological spring often mobilized in Trump – against an Anthropic which did not comply, at least publicly, with the instructions of the White House and the Pentagon. Personally, I see in this conflictual relationship between Anthropic and the White House less the scripted opposition between a pacifist approach and a militarist vision of AI than a battle around risk management of a technology which is now essential in the security field.
According to the American press, it was Amazon which detected security vulnerabilities and alerted the White House. What does this say about the relationship between the Trump administration and the tech giants?
Amazon, one of the first shareholders of Anthropic, a company largely dependent on AWS for the storage of its data, was not a priori the intended culprit! The affair is however revealing of the raw balance of power which is unfolding in the United States, in my opinion first and foremost between the tech giants themselves. Before Trump’s decision, Anthropic’s valuation was higher than that of its great rival OpenAI, which won the colossal contract with the Pentagon a few months ago. Amazon has also developed superlative ambitions in AI applied to industrial design. Let us also not forget the essential character of Google in the sector, nor the locking of the computing capacity market by Nvidia.Â
All these actors covet the gaze of the White House to increase their influence, and Trump does not hesitate to cajole them as long as they serve his ideological agenda and his unbridled business dealings. But it works both ways. The more the private sector swallows up the collection and processing of AI data, the more the State becomes dependent on it, which, in the United States, has become evident in the legal domain: public security, intelligence, military operations… The Anthropic affair should undoubtedly be read in light of this. of a political authority that stands up to a tech player announced as all-powerful. There will likely be repercussions for what remains of American soft power, since this affair allows – again in 2026! – a large inventory of our critical dependencies on the United States.
The boss of Anthropic, Dario Amodei, himself warned of the risks concerning his product. Arthur Mensch denounced “fear marketing” aimed at “solidifying monopolies”. Today, is Amodei’s strategy backfiring?
We can assume so! For years, Dario Amodei has built the Anthropic brand on a message, which can be summarized as follows: “Our AI is so powerful that it can become a threat. We must therefore regulate it, perhaps even slow it down. » This speech was something of a narrative, intended to position Anthropic as a model student of Silicon Valley, while impressing investors. It worked, based on the company’s sky-high valuations, but a story is never neutral. The calendar adds a degree of brutality to the affair: Anthropic is preparing its IPO, valued at more than 1,000 billion dollars. On the one hand, we must reassure the markets, on the other, the White House is cutting off part of the planet. European clients, the big banks who were testing its models, everything found themselves frozen overnight.
[…]
>> Find this interview in full on the Libération website.
