Brussels gains access to Mythos 5, Anthropic's most sensitive AI, after months of negotiations
The European Union Agency for Cybersecurity (ENISA) is already testing the model, which has been subject to very restricted access and, until now, has been primarily limited to the United States and a handful of partners. The agreement comes later than the June announcement and does not include the most recent version, Mythos 5.1.
Anthropic logo. File photo
The European Union has gained access to Mythos 5, Anthropic's most powerful and restricted artificial intelligence model, following several months of talks with the U.S. company, the European Commission confirmed on Thursday.
The European Union Agency for Cybersecurity (ENISA) is already testing the system, designed to assess cyber vulnerabilities and offensive capabilities in cyberspace, said Thomas Regnier, spokesperson for the Commission's Digital Affairs Directorate.
"Following our constructive dialogue with Anthropic, we can confirm that ENISA has gained access to Mythos 5 and is currently testing it," Regnier said.
Technology
Frantic advance of AI sets off alarms: Resignations, agent hacks, official warnings, possibility of human extinction
Emmanuel Alejandro Rondón
Why Mythos matters
The key difference is the access policy: Mythos is provided to a select group, with fewer restrictions than Fable 5. It is designed for cyber defense and penetration testing, and as such can also be used to design more sophisticated cyberattacks.
Politics
The National Security Agency warns that Chinese companies are extracting capabilities from US AI models
Williams Perdomo
Limited and delayed access
According to AFP, the agreement comes more than three months after Anthropic announced in June that the EU would be part of the so-called Glasswing Project, a select group of governments and companies authorized to test Mythos for its cybersecurity risks.
However, the success is only partial: the Commission acknowledged that ENISA will not be able to test the latest iteration of the model, Mythos 5.1, released just days ago, which incorporates even more advanced capabilities in code generation and vulnerability analysis.
Anthropic has so far maintained very restricted access to Mythos, reserved almost exclusively for U.S. agencies and a handful of international partners, due to its potential to identify and exploit critical flaws in computer systems.
LATEST
