The European Central Bank has urgently convened euro zone banks to assess cyber security risks triggered by Anthropic's advanced AI model, Claude Mythos.

On May 25, blockchain outlet BeInCrypto reported that the ECB plans to focus at a May 26 meeting on how quickly banks apply security patches and their systems for responding to AI-based attacks.

The core of the warning is that advanced AI is rapidly improving the speed and precision with which it finds security vulnerabilities. Frank Elderson, vice chair of the ECB's supervisory board, said banks need to deploy software patches much faster than before. He said attackers can now reverse-engineer patch contents within 30 minutes and that banks need to speed up their response.

Elderson stressed that cyber security tasks discussed with the banking sector for years remain valid, but processing speed must increase as AI advances. In an interview with the Financial Times, he said, "Cybersecurity issues overall are still important, but with advances in AI they must be handled more quickly." He added, "In musical terms, andante (a bit slow) is not enough, we need to go presto (very fast)."

The backdrop drawing attention from markets and supervisors is the performance of the Claude Mythos Preview. Anthropic released the model in April under a limited programme called Project Glasswing. The UK AI Security Institute said Mythos Preview passed 73 percent of expert-level Capture The Flag (CTF) security challenges. No AI model had achieved that level before April 2025.

The impact has also been confirmed in the software industry. Mozilla applied 271 security patches to Firefox150 to reflect vulnerabilities found by Mythos. That figure far exceeds the previous Opus4.6 result. For the financial sector, advances in vulnerability detection can strengthen defences, but they are also a burden because they can give attackers stronger tools.

The ECB currently supervises 111 large banks in the euro zone. However, most European banks are not participating in Project Glasswing and therefore do not have direct access to front-line AI models like Mythos. Elderson said he wants U.S. institutions attending the May 26 meeting to share test results and experience with euro zone banks. He called the access gap "regrettable" but drew a line that it was no reason to delay a response.

The key in this situation is whether the financial sector can keep up with the speed at which advanced AI reveals new vulnerabilities. If banks fail to shorten patch deployment cycles and cut initial response times after vulnerabilities are disclosed, the system for protecting customer funds could also face greater strain. The ECB's emphasis at the meeting is closer to response speed and an operational shift than to the technology itself. As AI is changing the security environment faster, the banking sector's cyber response standards are increasingly likely to be reset as well.