Anthropic opens Mythos: EU cyber agency to gain access
After long urging, the EU is to gain access to Anthropic's cyber AI Mythos. However, the exact conditions and timing are still open.
Seat of ENISA in Heraklion, Crete.
(Image: ENISA)
Anthropic is to grant the EU cybersecurity agency ENISA access to its cyber AI Mythos and include the authority in “Project Glasswing”, reports Bloomberg. Initially, primarily US partners had access, including government agencies, companies, and several banks. The British AI Security Institute was also allowed to test Mythos. The goal is to secure critical software with the help of Mythos before comparable capabilities become more widely available. According to Bloomberg, ENISA is the first EU authority with access to Mythos.
A spokesperson for the EU cybersecurity agency ENISA confirmed to the Financial Times that discussions about access to Mythos were underway, but emphasized that the conditions were still being negotiated. This also includes how far the US company's access to EU systems extends when European bodies use Mythos, the newspaper reports, citing people familiar with the discussions.
The EU Commission also confirmed discussions with Anthropic. The Commission has had several productive meetings with the company and welcomes the latest developments regarding possible access, a spokesperson told Bloomberg, without specifying a timeline. Representatives of the EU Commission traveled to San Francisco last week, according to reports, to negotiate with Anthropic about inclusion in “Project Glasswing”.
Videos by heise
The cyber AI threat doesn't end with Mythos
The EU has been pushing for access since the introduction of Mythos in April, but for a long time without success. Anthropic repeatedly cited the model's exceptional performance in finding security vulnerabilities and the associated potential for misuse when granting limited access.
Anthropic indicated a broader release last week, but spoke of “Mythos-class models.” The version currently being tested as part of Project Glasswing, “Claude Mythos Preview,” is not intended for general release, according to Anthropic.
Meanwhile, European banks are looking for their own ways to prepare for the cyber AI threat. For example, BNP Paribas expanded its partnership with the Parisian AI startup Mistral AI and is preparing for a new generation of powerful cyber AI models. After all, even if the EU gains access to Mythos, the dangers of AI-powered cyberattacks will not be averted.
(vbr)
