US clears Anthropic Fable 5 after safety fixes take hold

Anthropic Fable 5 is returning to global users after US officials lifted access curbs tied to cybersecurity safeguards and model oversight.

Anthropic PBC said Tuesday that the Commerce Department had notified the company it was removing restrictions on both Fable 5 and Mythos 5. The decision follows talks between the San Francisco-based AI company and officials over safety concerns raised by the Trump administration.

Commerce curbs give way

The dispute began with a June 12 private letter from the Commerce Department requiring Anthropic to seek US approval before foreign nationals could use Fable 5 or Mythos 5, even if they were physically in the United States. Anthropic responded by disabling both models while it worked through the government’s concerns.

Anthropic said the latest notice removed the remaining curbs on Fable 5 and Mythos 5, after some limits on Mythos had already been eased on June 26. The sequence turned a private export-control action into an early test of how Washington may handle access to advanced AI systems with cybersecurity uses.

Cyber filters reshape release

Fable 5 is coming back with additional classifiers that Anthropic said are designed to identify and block a broader range of cybersecurity tasks. The company framed the relaunch as both a product decision and a safety response to the government review.

Anthropic is also developing a framework to study jailbreaks and evaluate model responses with Amazon.com Inc., Microsoft Corp., Alphabet Inc.’s Google and other partners. The work is tied to Project Glasswing, a group of roughly 200 organizations that Anthropic has cleared to use Mythos 5 to search for cyber vulnerabilities.

The company said Fable 5 access would open to users globally on its own platforms beginning Wednesday. It also said availability through cloud services run by Amazon, Google and Microsoft would follow as quickly as the companies can implement it.

Mythos access stays narrower

Mythos 5 remains on a more controlled track than Fable 5. Anthropic said access has been restored for some US organizations, and that it will keep working with Washington to broaden eligibility for additional domestic and international Project Glasswing members.

The distinction matters because Anthropic is treating the two models differently: Fable 5 is intended for wider release, while Mythos 5 is tied more closely to vulnerability-hunting work. For the AI sector, the case shows that national-security review can affect not only model design, but also who can use a system and through which platform.

Rules debate widens

Anthropic said its cooperation with the government and industry partners could help shape rules for other frontier AI developers. The company urged regulations that are "applied equally across frontier model developers" and called for a "durable, transparent" process for government involvement in model releases.

If the new classifiers satisfy officials over time, the global rollout gives Anthropic a clearer path to distribute Fable 5 while giving cloud partners more predictable access terms. At the macro level, smoother approved releases could support cross-border AI services without immediately widening technology-trade friction.

If Washington instead relies on private, case-by-case letters, Anthropic could face slower launches and competitors could confront uneven compliance risks. If misuse concerns resurface, access could narrow again, pushing cloud platforms toward stricter customer screening and slowing adoption of advanced AI tools in cybersecurity-sensitive sectors.