Mythos Preview carve-out complicates Anthropic's US shutdown

A new wrinkle has opened in Anthropic's Fable/Mythos shutdown: the public and upgrade paths were cut off, but Mythos Preview appears to remain available to some early Project Glasswing users.

Reuters reported, citing Bloomberg News, that some firms selected by Anthropic to test Mythos preserved access to a preview version even after a U.S. government order led to the shutdown of other versions of the model family.1 The report matters because it separates three things that can easily get blurred: Fable 5, Mythos 5, and the older Mythos Preview channel inside Project Glasswing.

According to Reuters, Mythos Preview had been limited to about 200 organizations, including the U.S. government, through Anthropic's Glasswing program after the model identified thousands of software vulnerabilities.1 Reuters said Dragos and Cisco Systems confirmed to Bloomberg that they retained access to Mythos Preview, while Anthropic, Cisco, and Dragos were not immediately available for Reuters comment outside business hours.1

That access does not mean the broader shutdown is over. Anthropic's own statement on Jun. 12 said the U.S. government directed it to suspend all access to Fable 5 and Mythos 5 by any foreign national, including foreign national Anthropic employees.2 Anthropic said the practical result was that it had to disable Fable 5 and Mythos 5 for all customers to ensure compliance, while access to other Anthropic models would not be affected.2

The open question is whether Mythos Preview sits outside the directive's operative scope, or whether access is being handled through a narrower controlled program while the company and government sort out the order. Reuters said it was not immediately clear how Anthropic was determining access to individual Glasswing members.1

Anthropic's Jun. 9 launch post described Fable 5 as a Mythos-class model made safe for general use, with safeguards that route some risky requests to Claude Opus 4.8 instead.3

In the same post, Anthropic said Mythos 5 was for a small group of cyberdefenders and infrastructure providers, had safeguards lifted in some areas, and would initially deploy through Project Glasswing as an upgrade to Claude Mythos Preview.3 That history makes the Reuters/Bloomberg report more than an availability footnote. If some organizations can still use Mythos Preview while Fable 5 and Mythos 5 remain disabled, Project Glasswing becomes the live test case for whether the government will tolerate highly capable cyber-focused AI under a vetted-access model.

The carve-out, if it holds, is also uneven. Reuters said the European Union cybersecurity agency Enisa, which had been invited to join Glasswing before the U.S. block, was told it would no longer be given access.1 For enterprise buyers and security teams, that is the operational issue: access may now depend not only on model tier and customer trust status, but also on jurisdiction and licensing treatment.

Bloomberg separately reported that Anthropic co-founder Jack Clark said the company is in daily discussions with the Trump administration to resolve security concerns around Fable 5 and Mythos 5.4 Clark declined to give details on the talks, but Bloomberg said he framed Anthropic and other AI companies as trying to address national security challenges raised by rapid AI advances.4

Three checkpoints now matter for this feed:

Until one of those happens, the confirmed picture is narrow: some early Mythos Preview users reportedly still have access; Fable 5 and Mythos 5 remain under Anthropic's announced shutdown; and the access rules for Project Glasswing are still not transparent enough for customers to plan around.