Washington Suspends Mythos 5: A Turning Point for AI

Washington Suspends Mythos 5: A Turning Point for Artificial Intelligence

Dario Amodei, CEO of Anthropic

The U.S. government recently made a decision that has shaken the tech world: it ordered Anthropic to immediately suspend access to its most advanced artificial intelligence models, Mythos 5 and Fable 5. This directive is based on national security concerns and the export control powers held by Washington. This intervention marks a turning point in how the United States manages cutting-edge technologies.

Anthropic quickly responded by announcing the global deactivation of these two models. The company explained that this decision followed the identification of a potential mechanism to bypass the safeguards of Fable 5. However, Anthropic contests the severity of the concerns raised by U.S. authorities, asserting that the capabilities highlighted by the government are already available in other publicly accessible models. This situation underscores the tensions between tech companies and state regulators.

This intervention from Washington is unprecedented. Until now, the U.S. government had primarily focused on controlling infrastructures such as semiconductors, production equipment, and computing capabilities. With Mythos 5, it is the first time that a model itself has been directly targeted by a restriction. This decision could mark the beginning of a new phase in the governance of artificial intelligence, forcing laboratories to rethink their relationships with states.

Mythos 5: A Historic Precedent for Frontier Models

The U.S. government's abrupt intervention has caught observers accustomed to discussions about the security of frontier models off guard. This precedent is significant: while the industry has already experienced parliamentary inquiries and export restrictions on technologies, never before has an immediate withdrawal of a frontier model been demanded by a government.

The implications of this decision extend far beyond the specific case of Anthropic. What is at stake is the very principle of state intervention in the dissemination of capabilities produced by artificial intelligence systems. For the first time, a state asserts its right to intervene directly in this domain.

Towards an Operating License for AI Laboratories?

The Mythos affair could accelerate the rapprochement between major American laboratories and the federal government. For several years, players like Anthropic, OpenAI, and Google DeepMind have collaborated with government agencies through red teaming exercises and national security partnerships. Until now, these collaborations primarily aimed to manage risks. The Mythos episode could transform these relationships into a form of implicit license to operate.

The suspension of Mythos 5 reveals a new reality: it is no longer just about developing the most powerful systems, but about proving that they can be used without conflicting with U.S. strategic interests. This evolution brings AI laboratories closer to industries like defense or aerospace, where collaboration with the state is crucial.

European Dependence and Digital Sovereignty

For Europe, the Mythos affair sends a strong signal. The debate on digital sovereignty has so far focused on infrastructures and data localization. However, Mythos highlights a deeper dependence: regardless of where users or data are located, a critical model under U.S. jurisdiction remains subject to Washington's decisions.

This situation underscores the importance for Europe to develop its own capabilities in AI. As these capabilities become strategic elements of economic competitiveness, their control becomes as crucial as that of energy or telecommunications.

Open Source: A Strengthened Alternative?

Ironically, Washington's attempt to control closed models could enhance the appeal of open-source models. If a centralized model can be suspended, an open-source model is much harder to retract once distributed. The restriction of Mythos adds a geopolitical dimension to this debate.

For advocates of open source, this episode demonstrates that a critical cognitive infrastructure cannot depend on a few companies subject to the decisions of a single state. For governments, this reinforces the fear of uncontrolled proliferation of sensitive capabilities.

The Future Markets of AI

Considering Mythos as merely a cybersecurity-related case would be a mistake. Cybersecurity is likely just the first area where the capabilities of models reach a strategic value sufficient to attract the attention of states. Tomorrow, similar debates could emerge in sectors like drug discovery, aerospace, or defense.

The stakes are not only technological but also economic. The value of a model could soon be determined by its ability to accelerate a specific economic or strategic activity, marking the entry into the era of “capability models.”

Sovereign Risk: A New Variable for Investors

This evolution could also change how markets evaluate AI laboratories. Until now, discussions focused on revenues, future margins, and access to compute. The Mythos affair introduces sovereign risk as a new variable. The more strategic the models become, the more their value depends on their ability to remain authorized.

Frontier laboratories could be perceived as companies operating in regulated sectors, where the relationship with the state is an essential economic asset.

In conclusion, the question is no longer whether Mythos 5 will be reactivated, but how the ecosystem will react to this precedent. The suspension of a frontier model by a government could transform industrial strategies and technological choices across an entire industry. Initial reactions should not be long in coming.

When asked about Anthropic's decision, Anne Le Henanff, the minister in charge of Digital Affairs, believes this episode confirms the entry of AI into a new phase of strategic competition. She sees it as a justification for European efforts to strengthen their technological autonomy, highlighting France and Europe's advantages, such as abundant and competitive electricity, to host data centers. She also emphasized that the U.S. decision to restrict access to Anthropic's latest AI models to American users alone sends a clear message about the use of strategic technologies as instruments of power. This reality reinforces Europe's choice to build its technological autonomy.