The US government ordered Anthropic late Friday to immediately suspend access to Mythos 5 and Fable 5, its most advanced cybersecurity models. The directive is based on national security considerations and the export control powers available to Washington.
Anthropic announced a few hours later the global deactivation of both models. The company claims that the decision follows the identification of a possible mechanism for circumventing the safeguards of Fable 5. The laboratory, however, disputes the seriousness of the elements put forward by the American authorities and maintains that the capabilities put forward are already available in other publicly accessible models.
This intervention is unprecedented, since the start of the generative AI cycle, Washington has mainly focused on controlling infrastructure: semiconductors, production equipment and computing capacities. With Mythos 5, it is directly a model which is subject to a restriction. A decision which could mark the opening of a new phase in the governance of artificial intelligence and force laboratories to rethink their relations with States.
Mythos 5, first recall of a frontier model
The brutality of the intervention surprised everyone who has been following the debates on the safety of frontier models for years. The precedent is considerable, the industry has already experienced parliamentary inquiries, restrictions on the export of sensitive technologies and debates over the responsibility of model suppliers. But never has a government demanded the immediate withdrawal of a border model already deployed.
The scope of the decision goes far beyond the case of Anthropic; what counts now is not only the model concerned but the established principle. For the first time, a State affirms that it can intervene directly in the diffusion of a capacity produced by an artificial intelligence system.
Has Washington just imposed an operating license on AI laboratories?
One of the most underestimated effects of the Mythos affair could be the acceleration of the rapprochement between large American laboratories and the federal state.
For several years now, Anthropic, OpenAI and Google DeepMind have been multiplying red teaming exercises with government agencies, partnerships with national security organizations and supervision systems intended to supervise the most advanced models. Until now, these mechanisms were mainly related to risk management. Episode Mythos could give them another function and gain some form of implicit license to operate.
Because behind the suspension of the Anthropic model a new reality is emerging. The question is no longer just to build the most efficient systems, but to demonstrate that they can be deployed without coming into conflict with American strategic interests.
This development is gradually bringing AI laboratories closer to industries already familiar with this type of relationship with the State: defense, nuclear or aerospace. Except for years, Silicon Valley cultivated the idea that it could move faster than institutions, part of its future could now depend on its ability to move forward with them.
Europe discovers a deeper dependence which affects its development capacities
For Europe the signal is also very strong. The European debate on digital sovereignty has focused on infrastructure, data localization, secure clouds, etc. Mythos reveals a dependence of another nature because basically, no matter where the user is, no matter where the company is located, no matter even where the data is located, when a critical model is developed, hosted and governed under American jurisdiction, its access remains subject to American decision alone.
This is probably the most important lesson of this episode for European decision-makers. Europe is thus discovering that its dependence is no longer only on digital infrastructures but on cognitive capacities. And as these capabilities become strategic components of economic competitiveness, the question of their control takes on an importance comparable to that which energy, telecommunications or semiconductors may have had.
The argument for European models is no longer based solely on political sovereignty. It gradually becomes a subject of operational continuity.
By wanting to control closed models, Washington could strengthen open source
The paradox of this affair is that it could produce exactly the opposite effect to that sought, because the more governments seek to regulate proprietary models, the more attractive open models become.
If a model hosted by a centralized actor can be suspended, an Open-weight model is much more difficult to remove from circulation once distributed. This reality was already at the heart of the debate between closed and open source models, and the restriction of Mythos gives it a geopolitical dimension.
If for supporters of openness, it constitutes an additional demonstration that a critical cognitive infrastructure cannot depend exclusively on a few companies subject to the decisions of a single State, for governments, on the contrary, it reinforces the fear of an uncontrolled proliferation of sensitive capacities.
Behind the technical debate lies a broader question: will the future of AI be organized around licensed and supervised models, or around widely distributed and difficult-to-control capabilities?
The real topic: future AI markets
The mistake would be to consider Mythos as an isolated case linked to cybersecurity.
Cybersecurity is likely only the first area in which model-produced capabilities reach a level of strategic value sufficient to attract state attention. Tomorrow, the same debates will emerge around drug discovery, materials research, energy design, aerospace, defense and even intelligence.
The issue is therefore not only technological, but economic. Until now the value of a model has been measured through its general performance and another logic is gradually emerging, namely that the value of a model could soon be determined by its capacity to accelerate a specific economic or strategic activity.
The industry is thus entering the era of “capability models”, systems whose main resource is no longer general intelligence but the operational advantage they produce in a given domain.
Should investors start pricing sovereign risk?
This development could also change the way markets evaluate large laboratories.
After the debates on revenues, future margins, access to computing, talents or market shares. The Mythos affair introduces a new variable with sovereign risk. The more strategic models become, the more their value depends on their ability to remain authorized.
This is a major change in perspective, frontier laboratories were until now analyzed as the future software giants. They could gradually be seen as companies operating in regulated sectors where the relationship with the State constitutes an economic asset in its own right.
To conclude, the real question is now no longer whether Mythos 5 will be reactivated, but how the entire ecosystem will react to this precedent. Because when a government demonstrates that it can suspend a frontier model, it is the industrial strategies, investment flows and technological choices of an entire industry that are called upon to evolve. The first answers or reactions should not be long in coming.
