Bundesbank Chief: Anthropic Must Give Equal Access to Its Mythos Model

Joachim Nagel, president of the Deutsche Bundesbank, has called on Anthropic to give equal access to its Mythos model to all organizations it directly affects. In his words, an objective and independent assessment of both the model's potential and its risks is possible only if a level playing field, meaning equal conditions for all interested parties, is respected.

The Bundesbank chief's position is more than a political statement. It reflects deep concern among European financial regulators about how quickly advanced AI systems are entering critical sectors without an adequate oversight mechanism. If regulators, central banks and affected organizations cannot access the model on the same terms as large commercial customers, an information asymmetry arises. That is what worries the German central banker most.

Mythos is one of the most powerful models Anthropic has introduced in 2026. The company positions it as an extended-reasoning system capable of handling complex professional tasks: financial analysis, legal review, scientific research and organization-level strategic planning. That use profile maps directly onto the work of financial institutions, which is why Nagel's request looks systemic rather than incidental.

The European context matters here. The EU Artificial Intelligence Act (AI Act) requires developers of high-risk systems to undergo independent assessment and provide technical documentation. For models with the greatest potential impact, so-called GPAI models with systemic risks, the requirements are stricter still. However, a mechanism for regulators to access the models directly for independent testing has still not been standardized. This is precisely the question Nagel raises: compliance on paper is one thing, the ability to examine the system yourself is quite another.

Central banks find themselves in a dual role. On one hand, they are studying AI in their own work: forecasting macroeconomic indicators, supervising systemic risks and monitoring financial stability in real time. The European Central Bank and a number of national regulators have already launched pilot projects using large language models in internal processes. On the other hand, the same institutions must track how commercial banks and financial firms deploy external AI systems in lending, risk management and automated trading. Both tasks require direct access to the model, not just analysis of the developer's documentation.

Anthropic has not yet commented on Nagel's statement. The company traditionally stresses responsible development: Constitutional AI, red teaming with outside experts and its Responsible Scaling Policy are all part of its public safety positioning. However, declared openness and providing technical access to regulators are fundamentally different things. Controlled regulatory access to commercial AI models has not yet become an industry norm.

The Bundesbank president's statement signals that regulatory pressure on leading AI developers in Europe will keep growing. The gap between the pace at which new models reach the market and the ability to assess them independently is becoming a structural problem. When institutions such as the Bundesbank publicly set equal access as a condition, it creates a precedent and strengthens those who argue for tougher AI regulation in Europe's financial sector.