Roughly 200 companies that had been granted early access to an Anthropic AI model called Mythos have managed to keep that access even after a recent US government shutdown order, according to Bloomberg.
The companies reached the model through an effort referred to as Project Glasswing. According to Bloomberg, these firms were chosen early on by Anthropic PBC to test the Mythos model ahead of a wider release, and they preserved their access to the preview despite the shutdown order.
The situation appears to be a headache for Anthropic. A separate Bloomberg piece is headlined "Anthropic Has the Template for Fixing Its Mythos Nightmare," signaling that the rollout of the model has run into trouble the company is now working to address. The sources here do not spell out what that template is or exactly what went wrong.
The details available are limited. The sources confirm the approximate number of companies involved, the name of the model (Mythos), the name of the access program (Project Glasswing), the developer (Anthropic PBC), and the fact that a US government shutdown order was in play but did not cut off the testers' access.
Why it matters: when a government shutdown order fails to stop outside companies from using an unreleased AI system, it raises questions about who controls access to powerful new models and how effectively such orders can be enforced.