ECB convenes banks to fix flaws exposed by AI Models, FT says

The European Central Bank will urge banks to accelerate IT security upgrades after AI models like Anthropic’s Claude Mythos Preview exposed critical flaws, with ECB Executive Board member Frank Elderson warning that delays risk exploitation by malicious actors. The EU is negotiating with Anthropic for vulnerability testing access, while French startup Mistral AI offers an alternative solution for European banks." "article": "The European Central Bank (ECB) is convening a meeting with lenders on Tuesday to address cybersecurity risks amplified by advanced artificial intelligence models like Anthropic’s Claude Mythos Preview. ECB Executive Board member Frank Elderson emphasized the need for faster action, stating that existing vulnerabilities—long discussed with banks—must now be resolved urgently due to AI’s ability to identify flaws within minutes of patch releases. Elderson warned that European banks cannot use limited access to Mythos as an excuse for inaction, as malicious actors may soon exploit the technology. The ECB will press US banks with Mythos access to share insights with European counterparts, who currently lack direct testing opportunities. Anthropic disclosed last month that Mythos uncovered thousands of severe vulnerabilities across major operating systems and browsers. Meanwhile, the European Union is negotiating with Anthropic to grant EU entities access for vulnerability assessments, as revealed by European Commissioner Valdis Dombrovskis on May 4. French AI startup Mistral AI is also in talks with European banks to deploy its own security-flaw detection tool. Elderson stressed that AI-driven threats demand immediate software patching, as delays could leave financial systems exposed to rapid exploitation by cybercriminals.