AIster · The agentic age · Launches

Anthropic releases Fable 5, then the US forces it offline

A model held back for its hacking skill went public, then an export order pulled it

Key facts

Date

9 Jun 2026

Era

The agentic age

Interest

Launches

Threads

Model wars, Safety & governance, Labs & people

Who

Anthropic; US Commerce Department

Models

Claude Fable 5 (public), Claude Mythos 5 (restricted)

Price

$10 / $50 per million input / output tokens

Offline

Export controls 12–30 Jun 2026; Fable 5 back 1 Jul

Held back

On 7 April 2026 Anthropic announced Project Glasswing and a model it chose not to release to the public, Claude Mythos Preview. It said the model had found thousands of high-severity vulnerabilities across every major operating system and web browser, including a 27-year-old bug in OpenBSD. Access went to partners including Amazon Web Services, Apple, Cisco, CrowdStrike, Google, Microsoft, NVIDIA and the Linux Foundation, backed by up to $100 million in usage credits, so defenders could patch before such skills spread. Anthropic said models could now “surpass all but the most skilled humans at finding and exploiting software vulnerabilities”.

Released with guards

On 9 June Anthropic released Claude Fable 5 for general use alongside Claude Mythos 5, which stayed restricted. The two are the same underlying model. Fable 5 adds classifiers (automated filters) for cybersecurity, biology and chemistry, and for attempts at distillation, which means copying a model by training on its answers; when they fire, the answer comes from the less capable Claude Opus 4.8 instead. Anthropic said the safeguards triggered, on average, in fewer than 5% of sessions. Both cost $10 per million input tokens and $50 per million output tokens.

Pulled by export order

On 12 June the US government placed Fable 5 and Mythos 5 under export controls, barring access by foreign nationals inside or outside the United States. The trigger was a prompt, found by Amazon researchers, that got Fable 5 past its safeguards on software vulnerabilities. Because Anthropic could not check users’ nationality in real time, it suspended both models for all users worldwide. It then deployed new classifiers aimed at the technique. The controls were lifted on 30 June, and Fable 5 came back globally on 1 July.

Why it matters

Anthropic had judged Mythos too risky for public release in April, then shipped it in June with classifiers and a fallback model as extra safeguards. Within days the US government used export rules to take the commercial model offline, showing that access to a frontier model could be switched off by the state at short notice.

Questions people ask

What is the difference between Claude Fable 5 and Claude Mythos 5?

They are the same underlying model. Fable 5 is the public version, with classifiers covering cybersecurity, biology, chemistry and distillation that hand flagged requests to Claude Opus 4.8. Mythos 5 has some of those safeguards lifted and was released first to Project Glasswing partners, with select biology researchers to follow.

Why was Claude Fable 5 taken offline?

On 12 June 2026 the US government put Fable 5 and Mythos 5 under export controls after a report in which Amazon researchers found a way past Fable 5’s safeguards. Anthropic could not verify users’ nationality in real time, so it suspended both models worldwide until the controls were lifted on 30 June; Fable 5 returned on 1 July.

Sources

1. Anthropic, Project Glasswing: Securing critical software for the AI era, 2026-04-07
2. Anthropic, Claude Fable 5 and Claude Mythos 5, 2026-06-09
3. Anthropic, Redeploying Claude Fable 5, 2026-06-30