Issue #23 - June 26, 2026
- Mythos is a warning of a future weaponization of frontier models by governments - The DMA extends to US Cloud providers in a context of US/EU tension over trade deal
Sébastien Louradour
6/26/20264 min read
Mythos is a warning of a future weaponization of frontier models by governments
Over the past few weeks, the export ban of the latest model launched by Anthropic by the US has raised tremendous concern over all European capitals. Anthropic was relatively unknown by European policymakers in April when compared to its main competitor Open AI, but the release of Anthropic's Mythos model triggered immediate concern across Washington and spread quickly. With the potential to supposedly detect any kind of cybersecurity failures and breaches, many US Government voices quickly raised concern over the risk of seeing such a model in bad hands. Followed a few meetings with a select group of critical US companies with Anthropic to provide them with the model and so they can fix potential cybersecurity threats before it's too late to act.
A few weeks ago, Chris Olah, the co-founder of Anthropic, appeared with the Pope to join the concerns the Church has about uncontrolled use of AI. Within a few months, the company gained the status of being seen by some as the first to reach AGI-level capabilities, the grail of AI, which brought with it the attendant questions of existential risk the company has long flagged.
In this context, the conditions for a crisis were already in place, and when Anthropic finally decided to release the public version of Mythos, called Fable, it only took a few voices to raise enough concern among the US administration over a potential systemic risk to call for an export ban of the model, which as an effect forced the company to suspend access entirely on Fable. The reaction of the EU, interpreting this as a kill switch and a ban from Europeans, was partly right, but was also based on months of repeated efforts to prioritize a tech sovereignty agenda, which reinforced an existing political narrative around sovereignty. Even though Fable was technically shut down for anyone, Europe saw it as an additional proof point that AI sovereignty should be a priority to avoid depending on unilateral decisions made by foreign governments.
The start of the roll out of ID verification for Anthropic's users shows already that the company is prepping to avoid a complete shut down of its services and be able to comply with rules by knowing the identity and citizenship of its users. Right now, Anthropic says that "Identity verification helps prevent abuse, enforce usage policies, and comply with legal obligations". Now the question is whether the US administration ban of non-US citizens using high-end services of Anthropic is going to be enforced.
The G7 summit last week in France gathered the leaders of the main AI Labs and head of states, where G7 allies pushed for a 'trusted partner' scheme that would give close US allies privileged access to frontier models, but no formal agreement was made on the issue. While US AI Labs have been advocating for access to their most advanced models to US allies, the question of export control remains open, with the risk of ending decades of US soft power promoting open globalisation and a global internet with little or no frontiers. What the Mythos episode has already settled, regardless of what comes next, is that access to frontier AI is now a matter for governments, not companies, to decide.
The DMA extends to US Cloud providers in a context of US/EU tension over trade deal
From the start, the EU Digital Markets Act (DMA) has been a considerable issue for the biggest tech companies. Even a few weeks ago, Apple publicly blamed the regulation for slowing down access to its latest AI services in the EU. The US administration has also been very active criticising an indirect levy on the most successful US tech companies and has been constantly threatening of additional tariffs if the EU keeps enforcing the file on American firms. And despite the pressure both from US big tech and the US administration, the European Commission has announced this week in a preliminary finding, the extension of the scope of the DMA on cloud providers.
As for now, to be in scope of the DMA, a specific list of services of tech companies have to reach a certain number of EU users, for example, the App Store, Instagram, TikTok, and Booking are in the scope, and must apply certain mechanisms to keep the digital markets open and interoperable so new players are not de facto turned out because it's too hard to compete with already in place dominant players or "gate keepers". In practice interoperability has been hard to implement, to a point where tech companies had to make hard decisions to not face steep fines which has also led tech companies to challenge the designations. A recent decision by the EU General Court annulled the initial assessment that Facebook Marketplace was a gatekeeper.
It's now likely similar frictions appear for Cloud providers too. In a preliminary finding, the European Commission has announced AWS and Azure should be designated as gatekeepers. While they don't meet the quantitative thresholds, the Commission cites a list of qualitative arguments such as lock-in effects, high switching costs, as well as providing access to a portfolio of AI tools and partnerships.
The question of cloud sovereignty has already been raised in the EU Tech Sovereignty package presented a few weeks ago, and with the DMA designation of two US cloud providers, the compliance pressure is reaching a new high. The US Administration has been raising repeated concerns over what it sees as indirect levy on US companies, and has warned of potential tariff retaliation if the EU keeps targeting US companies with high enforcement requirements and fines. With this new announcement by the European Commission, it seems that Brussels is up for a fight.