A powerful new AI model designed to hunt deep software flaws is moving into a tightly controlled trial inside the European Union. The model, Mythos from Anthropic, has drawn attention because it is not just built to analyze code, but to uncover security weaknesses that conventional tools often miss.
That promise comes with an equally serious concern. Anthropic has said the model is highly effective at finding vulnerabilities, which means it could also be misused if access is not carefully limited.
A controlled first test inside the EU
The first EU body set to receive access is ENISA, the bloc’s cybersecurity agency. According to Bloomberg, European officials have already been in contact with Anthropic and even traveled to San Francisco to request access to the technology.
That access is being granted through Project Glasswing, Anthropic’s early-access program for a select group of organizations. The idea is to let trusted testers explore the model in a safer environment before any wider release is considered.
For the European Union, the trial offers a direct way to assess whether a frontier AI model can really expose weaknesses that remain hidden from standard security systems. It also gives regulators a closer look at how far such powerful tools should be allowed to go in cybersecurity work.
What makes Mythos different
Mythos stands out because it is described as capable of finding software vulnerabilities that are extremely difficult to detect. In practical terms, that could help security teams close gaps before attackers get to them.
The model is also said to be able to simulate multi-step cyberattacks. That ability could make it useful for governments and businesses that want to test how resilient their systems are against complex attack paths.
Used carefully, a tool like this could strengthen cyber defense in a meaningful way. As digital threats continue to rise, earlier detection of weak points has become one of the most valuable parts of security planning.
Why the same tool raises concern
The same capabilities that make Mythos attractive also explain why access is restricted. A model that can identify hard-to-find weaknesses may also provide harmful guidance if it falls into the wrong hands.
That is why Anthropic has kept the model out of public release. Since its preview in April, access has remained tightly controlled because the company has judged the misuse risk to be too high for open availability.
This is also why the EU trial matters beyond a single testing exercise. It gives ENISA and other policymakers a chance to study how a highly capable AI system behaves under controlled conditions, and how much oversight is needed before broader use can be considered.
Part of a wider, limited rollout
Project Glasswing has already involved several early testers before ENISA entered the picture. Those testers include regulators in the United States, several banks, and the AI Security Institute in the UK.
The EU’s participation marks an important expansion of that group. It also suggests that public authorities are beginning to view advanced AI for cybersecurity as something strategic, not just as another technical tool.
Talks between Anthropic and the European Commission are still ongoing. Officials have confirmed that the discussions are moving forward, although the details have not been made public.
That cautious approach reflects the broader challenge around frontier AI in security work. Tools like Mythos may help stop major attacks, but they also carry knowledge that could be repurposed for more sophisticated offensive operations.
For ENISA, the upcoming access could help define where that line should be drawn. The outcome may shape how Europe thinks about the safe use of high-end AI in cyber defense, especially as the pressure to respond to digital threats keeps growing.
Source: sundayguardianlive.com-
-
-
-
