Some firms chosen early on by Anthropic to test the Mythos AI model have preserved their access to a preview of the system, despite a US government order that led to the total shutdown of other versions, Bloomberg News reported on Thursday.
The Claude maker had limited Mythos Preview to about 200 organizations, including the US government, under its Glasswing program after the model identified thousands of software vulnerabilities.
A less powerful version of Mythos was publicly released, and then disabledunder Washington's export-control directives to suspend access for all foreign nationals due to national security concerns.
Firms such as Dragos and Cisco Systems confirmed to Bloomberg News that they had retained access to Mythos Preview. It was not immediately clear how Anthropic was determining access to individual Glasswing members, the Bloomberg report added.
Anthropic, Cisco, and Dragos didn't respond to Reuters' requests for comment.
European cybersecurity agency ENISA, which had been invited to join Glasswing ahead of the U.S. government's block, was informed on Friday it would no longer be given access, the report added.
ENISA said in a statement on Friday that it was continuing discussions with Anthropic on Mythos Preview access while also consulting other AI model providers and open-weight models to understand "today's operational pressures, tomorrow's strategic pivots, and the day-after-tomorrow's existential uncertainties."
IPO-bound Anthropic's relationship with the US government ruptured this year after it refused to allow the US military to use its AI models for domestic surveillance and fully autonomous weapons systems.
