EU probes Anthropic over cybersecurity risks tied to mythos ai model

Friday 17 April 2026 - 14:20

By: Dakir Madiha

EU probes Anthropic over cybersecurity risks tied to mythos ai model

European regulators have stepped up scrutiny of artificial intelligence risks after the release of Anthropic’s Claude Mythos Preview, with authorities seeking clarity on its potential to expose critical vulnerabilities in digital infrastructure. The European Commission confirmed it has opened direct discussions with Anthropic to assess the model’s implications, marking a coordinated response at the European level.

The European Central Bank is preparing to question banks about their readiness to handle emerging cyber threats linked to advanced AI systems. Supervisors are gathering information on the Mythos model and plan to raise the issue during routine oversight meetings with financial institutions. The move reflects concern that tools capable of identifying and exploiting software vulnerabilities could increase exposure for banks operating with legacy systems.

Anthropic has described Mythos as able to detect weaknesses across major operating systems and web browsers, raising alarm among regulators responsible for financial stability and cybersecurity. European authorities are particularly focused on the risk that such capabilities could be misused or accelerate the scale of cyberattacks targeting critical infrastructure.

The European Commission said initial talks with Anthropic have already taken place, with further meetings expected. Officials are seeking detailed information on the model’s capabilities, safeguards and deployment strategy. The engagement follows reports that most European cybersecurity agencies were not granted early access to the system, limiting their ability to test its behavior before release.

The rollout of Mythos has been tightly controlled under a restricted program involving a small group of major technology and financial firms. These include Apple, Google, Microsoft, Nvidia, Amazon Web Services, JPMorgan and CrowdStrike. Early findings reportedly include the discovery of long standing software flaws and thousands of previously unidentified vulnerabilities.

The United Kingdom is moving more quickly to integrate the technology into financial sector oversight. Bank of England is expected to brief major banks and insurers on the implications of the model, while Anthropic plans to expand access to British institutions in the coming days. The divergence in pace highlights differences in regulatory approaches across Europe.

Cyber risk has already been identified as a top supervisory priority for the ECB for the 2026 to 2028 period. Officials warn that the rapid evolution of AI driven tools is reshaping the threat landscape faster than traditional regulatory frameworks can adapt, forcing authorities to balance innovation with systemic risk control.