Anthropic has allowed a small group of organisations to retain access to its Mythos Preview artificial intelligence model following a US government order last week that forced the company to shut down other versions of the system, preserving a key cybersecurity research capability for selected users.
The restriction followed a US directive requiring Anthropic to block access to certain advanced AI models for foreign nationals unless a licence is obtained. In response, the company withdrew Fable 5 from public use and halted the rollout of Mythos 5, a less restricted version of the technology that had been undergoing wider testing.
According to people familiar with the matter cited by Bloomberg, some members of Project Glasswing, Anthropic’s cybersecurity research programme, continue to use Mythos Preview despite the broader shutdown. The initiative includes roughly 200 organisations authorised to test advanced AI systems for cyber vulnerabilities.
Several major companies remain involved. Jon Lavender, chief technology officer at Dragos, confirmed his company still has access to Mythos Preview. Cisco likewise confirmed that it has retained access to the model, while Amazon Web Services and JPMorgan Chase were among the programme’s earliest participants.
Bloomberg reported that Mythos Preview attracted attention during testing after identifying thousands of software vulnerabilities, including flaws across major operating systems and web browsers. Those capabilities prompted Anthropic to restrict access to the model in April, limiting availability to Project Glasswing members rather than releasing it more broadly.
The government order created uncertainty among participants over whether research versions of the model would remain available. The directive did not explicitly mention Mythos Preview, and Anthropic stated in a blog post only that “all other Anthropic models” would remain unaffected. The company did not respond to requests for comment on the status of the preview version.
Not all organisations connected to Project Glasswing have been spared from the restrictions. A spokesperson for the European Union Agency for Cybersecurity, known as Enisa, said the agency had been invited to join the programme but was informed on Friday that it would no longer receive access. A European Commission spokesperson said Enisa met with Anthropic on Thursday before the decision was communicated.
The differing treatment of participants suggests Anthropic is assessing eligibility on an organisation-by-organisation basis as it responds to US export controls, although the company has not disclosed the criteria being used.
