Anthropic moves to include EU’s ENISA in testing its most advanced AI system Mythos

Anthropic is set to grant the European Union’s cybersecurity agency ENISA access to its powerful AI model Mythos via the Project Glasswing programme, amid ongoing concerns over oversight and security in frontier AI systems.

Anthropic is preparing to give the European Union’s cyber security agency access to Mythos, its most advanced artificial intelligence system, after regulators in Brussels pressed for a closer look at a model designed to probe and exploit software weaknesses. According to people familiar with the talks, the company will allow the EU Agency for Cybersecurity, known as ENISA, to join its controlled-access Project Glasswing programme, which was created to let selected organisations test the model before any broader release.

The move follows weeks of limited engagement between Anthropic and the European Commission. Commission officials travelled to San Francisco on Thursday to seek access to the model, after earlier efforts to secure a review had not yielded results, people familiar with the discussions said. A commission spokesperson said talks with Anthropic were continuing and declined to give further details, while Anthropic did not comment.

Project Glasswing is Anthropic’s effort to share Mythos with a narrow circle of security and technology organisations so they can strengthen defences against the same techniques the model can use offensively. On its website, Anthropic says the initiative includes major companies such as Amazon Web Services, Apple, Cisco, Google, Microsoft and Palo Alto Networks, alongside more than 40 other organisations involved in critical software infrastructure. The company says it has committed up to $100 million in usage credits and $4 million in donations to open-source security groups to support the programme.

The restricted rollout reflects the model’s sensitivity. Anthropic has said Mythos is unusually strong at finding network vulnerabilities and could pose a serious risk if it were misused by criminals or other hostile actors. The model was first previewed in April, and since then access has been tightly controlled. Early users reportedly included US authorities, some American financial firms such as Goldman Sachs, and the UK’s AI Security Institute, which was allowed to test the system soon after launch.

The EU’s effort to get a seat at the table also speaks to a broader concern in Europe that public institutions should not be left behind as frontier AI systems become more capable of offensive cyber work. A report by iSec News in April said European regulators had been largely shut out of early access to Mythos, raising questions about oversight, security and sovereignty. Anthropic’s latest decision to include ENISA suggests those concerns may now be influencing how the company balances caution with external scrutiny.

Source Reference Map

Inspired by headline at: ^[1]

Sources by paragraph:

Source: Fuse Wire Services