German banks assess risks linked to Claude Mythos, Anthropic’s AI model

German banks are already weighing up the risks associated with Claude Mythos, Anthropic’s AI model that is particularly strong at finding security weaknesses. Used responsibly, it could help banks and online services harden their software; in the wrong hands, comparable capability could also be turned into a powerful tool for cyber-attacks.

Why Claude Mythos worries German banks

Germany is now paying close attention to Claude Mythos, Anthropic’s AI model designed to uncover software vulnerabilities-something that, if misused, could trigger a fresh wave of cyber incidents. Reuters reported this week that German banks, along with public authorities, have begun assessing the threat profile linked to the model.

Kolja Gabriel, a board member of the German Banking Industry Committee, said the association is consulting banks’ cybersecurity specialists and the Ministry of Finance as part of this work.

“Mythos is being used in a controlled manner by IT security companies to fix potential vulnerabilities as quickly as possible. We are expecting a series of software updates soon and are closely monitoring how the situation develops,” he added.

Germany’s financial regulator, BaFin, has likewise indicated that financial institutions should brace for the near-term discovery of new security gaps that will need to be patched quickly.

Project Glasswing: how Anthropic is containing Claude Mythos

Claude Mythos was formally unveiled in April, but Anthropic has not released it publicly. Instead, the company launched Project Glasswing, which aims to secure critical software by giving defenders a head start.

Under this initiative, the AI lab is working with major players including Amazon, Apple, Google, Microsoft and Nvidia, as well as JP Morgan Chase. Reuters notes that, before the project was announced, Claude Mythos had already identified thousands of very severe vulnerabilities, including flaws affecting widely used operating systems and popular web browsers.

What about Europe and France?

At European Union level, the European Central Bank (ECB) is already moving. According to Reuters, it has been questioning European banks about how prepared they are for this new category of risk. The European Commission, for its part, has already welcomed Anthropic’s decision not to release its Mythos AI model immediately.

“Given the large-scale cyber risk posed by this model, we welcome this gradual deployment,” a Brussels representative told Politico.

So far, there has been no public comment in France. Even so, given the scale of the risk-and the attention Claude Mythos is attracting-it is possible that discussions are already taking place behind the scenes. Anthropic also warns that other organisations could soon develop similar models. “Given the pace of AI progress, it will not be long before such capabilities proliferate, potentially beyond actors who commit to deploying them safely. The repercussions-for economies, public safety and national security-could be serious,” the company writes.

What we think

Anthropic’s messaging implies that a race against the clock is already under way. On its account, a model with capabilities comparable to Claude Mythos could end up accessible to malicious actors sooner or later. The objective, then, is to ensure critical software is ready before that point arrives.