European Union Demands Anthropic Access to Claude Mythos for Cybersecurity Tests

The European Union finds itself at a critical juncture in its negotiations with Anthropic, the company behind the Claude Mythos artificial intelligence model. Dialogue between the bloc and the US-based firm has stalled, raising concerns in Brussels about the cybersecurity of its companies and, particularly, its banking system. According to Bloomberg reports, the EU has made no significant progress in convincing Anthropic to allow Claude Mythos to be tested against European financial infrastructures.

Anthropic, led by Dario Amodei, has so far refused to share detailed information with European authorities regarding the potential threats its AI model could unleash. This reluctance has caused frustration, especially considering the importance of global financial stability. Carlos Cuerpo, Spain's Minister of Economy, recently acknowledged that progress in discussions with Claude Mythos's representatives was limited, emphasizing that addressing the risks posed by this AI is an immediate and long-term priority.

The primary concern in Europe is that its banking system could be exposed to unknown vulnerabilities when faced with an advanced model like Claude Mythos. To date, Anthropic has only granted access to its AI to a select group of companies and government agencies in the United States. The EU considers this geographical limitation unacceptable, given that the impact of a disruptive model can transcend borders and affect an interconnected financial system.

Financial stability is global. The impact that a disruptive model like Mythos, or others that will follow, may have on banks in the UK, Canada, and Europe can also extend to the United States through an interconnected system.

Despite European insistence, Anthropic has remained firm in its refusal to offer broad access to Claude Mythos's capabilities. However, the bloc assures that its efforts to gain access will continue. The company has indicated it will share findings from Claude Mythos with the Financial Stability Board, as reported by The Guardian. While this doesn't fully address the EU's demands for preparing its banking system against potential cybersecurity threats, it is seen as an encouraging step.

In the face of Anthropic's lack of cooperation, OpenAI has emerged as a potential ally for Europe. Sam Altman's company plans to launch a specialized version, named GPT-5.5-Cyber, specifically designed to address cybersecurity challenges. This new model will be made available to companies, governments, and authorities in Europe, offering an alternative to assess and strengthen digital defenses against emerging AI threats.

The situation highlights the growing tension between the need for innovation in artificial intelligence and the urgency of ensuring the security and stability of critical infrastructures. While models like Claude Mythos promise significant advancements, the lack of transparency and collaboration from their developers raises serious questions about how to manage the associated risks.

The European Union faces the challenge of regulating a rapidly evolving technology, seeking a balance between fostering innovation and protecting its citizens and economies. Anthropic's stance, while understandable from an intellectual property and model security perspective, creates a void that other companies, like OpenAI, appear ready to fill.

The future of financial cybersecurity in Europe will largely depend on regulators' ability to secure necessary cooperation from major AI companies and on the effectiveness of alternative solutions that emerge to mitigate the risks inherent in these powerful models.