Some companies selected by Anthropic to test its Mythos AI model have retained access to an earlier preview version despite a US government order that forced the company to shut down other versions.

Cisco and cybersecurity firm Dragos confirmed they still have access to Mythos Preview through Project Glasswing, according to a Bloomberg report.

Project Glasswing includes roughly 200 banks, technology companies, government agencies and other organizations approved by Anthropic to use the system for defensive cybersecurity work.

Anthropic limited access to Mythos Preview after the model identified thousands of software vulnerabilities across major operating systems and web browsers.

The company later released Fable 5 to the public with restrictions designed to prevent it from performing sensitive cybersecurity tasks. It also planned to roll out Mythos 5 to selected Project Glasswing members.

Anthropic abruptly disabled both versions last week after the US government ordered it to block access to Fable 5 and Mythos 5 for all foreign nationals unless they obtained a license.

The order did not explicitly address Mythos Preview, leaving some members uncertain over whether their access would continue.

Anthropic said in its announcement that all of its other models would remain available but did not directly clarify the status of the preview system.

Access has not been preserved across the entire program. The European Union cybersecurity agency Enisa was invited to join Project Glasswing before the government order but was later told it would no longer receive access.

It remains unclear how Anthropic is determining which organizations can continue using Mythos Preview.

The continued access gives selected companies a way to keep testing defensive tools against increasingly capable AI systems while Anthropic works through the restrictions affecting its newer models.