Claude Mythos Explained: Anthropic's Restricted AI Model
Updated 2026-10-12
Short answer: Claude Mythos is a series of Anthropic AI models whose cybersecurity abilities – finding and exploiting software vulnerabilities – led Anthropic to keep it out of general release. The first version, Claude Mythos Preview, was disclosed on April 7, 2026 and given only to vetted organizations through Project Glasswing. In June 2026 Anthropic released Claude Fable 5, a "Mythos-class" model with extra safeguards, for general use; the current versions are Mythos 5.1 and Fable 5.1 (September 1, 2026). Ordinary users get Fable, not Mythos.
Below is the timeline, the difference between Mythos and Fable, and what the security claims do – and don't – show. All dates come from Anthropic announcements as reported by major outlets and summarized on Wikipedia (checked October 12, 2026).
Claude Mythos timeline
| Date (2026) | What happened |
|---|---|
| March 26 | Leaked draft blog posts reveal a model called Claude Mythos; Anthropic confirms it is in development |
| April 7 | Anthropic discloses Claude Mythos Preview and launches Project Glasswing; no public release |
| May | Glasswing expands from about 50 to about 200 organizations in power, water, healthcare, communications and hardware |
| May 28 | With Claude Opus 4.8, Anthropic says "Mythos-class" models will reach all customers once extra safeguards are ready |
| June 2 | Access widened to 150 organizations in more than 15 countries |
| June 9 | Claude Mythos 5 (restricted) and Claude Fable 5 (with safeguards) released |
| June 12–30 | A US government letter forces Anthropic to pause access for non-US nationals; restrictions lifted June 30 and access restored July 1 |
| July 20 | Fable 5 becomes a standard feature of Anthropic's higher-tier subscription plans |
| September 1 | Claude Mythos 5.1 and Claude Fable 5.1 released |
| October 6 | Mythos 5.1 opened to users approved for the updated Cyber Verification Program |
Mythos vs Fable: what's the difference?
According to Anthropic, Mythos 5 and Fable 5 are the same underlying model. The difference is the safeguards:
- Claude Fable is the version for general use. When its classifiers flag a request about cybersecurity, biology and chemistry, or model distillation, the answer is handed to the less capable Claude Opus instead.
- Claude Mythos is the restricted-access version with some of those safeguards lifted, for vetted security organizations and approved programs.
So if you use Claude through a paid plan and see a "Mythos-class" model, it is Fable. Mythos itself requires approval. Industry estimates reported by the Financial Times put Mythos at roughly 8 trillion parameters and Fable 5 at roughly 5 trillion; Anthropic has not published official sizes.
Later models inherit the same approach: Anthropic says Claude Opus 5.5 has biology and cybersecurity skills comparable to Mythos 5.1 and ships with safeguards similar to Fable 5.1. See also our guide to Claude Fable 5.5.
Why was Claude Mythos restricted?
Anthropic said Mythos could find serious vulnerabilities in critical software, and that releasing it openly would help attackers as much as defenders. Instead, Project Glasswing gave access to companies that maintain critical software – Anthropic named partners including Microsoft, Apple, Google, Amazon Web Services, the Linux Foundation, Cisco, Nvidia and Broadcom – so they could find and fix bugs first.
The announcement had effects well beyond tech. Hours after the reveal, the US Treasury Secretary and the Federal Reserve chair met financial executives about it, several large US banks began testing the model, and central banks and regulators in Canada, the UK, India, Japan and Australia held their own meetings.
How strong is Mythos really?
The headline claims are Anthropic's, and some have been challenged:
- Anthropic said Mythos found vulnerabilities in "every major operating system and every major web browser". Mozilla later said it had found and patched 271 Firefox vulnerabilities using Mythos Preview, and Glasswing's first 50 partners reported over 10,000 high or critical findings.
- An independent researcher pointed out that most successful exploit runs relied on the same two (now patched) bugs, and another test found that small open-source models could spot one of the headline vulnerabilities too.
- A security researcher reported in September that only about 10% of discovered flaws had been disclosed and under 1% fixed – the bottleneck is human validation and patching, not discovery.
The fair reading: Mythos is a very capable security model, but "AI found 10,000 bugs" is not the same as 10,000 fixed, confirmed vulnerabilities.
Can I use Claude Mythos?
- Most people: no direct access. Use Claude Fable 5.1 (or Opus 5.5) on Anthropic's higher-tier plans – it is the same model family with safeguards.
- Security teams: apply to Anthropic's Cyber Verification Program; approved users got access to Mythos 5.1 from October 6, 2026.
- Critical-infrastructure organizations: Project Glasswing is invitation-based.
Beware of websites and apps that claim to offer "free Claude Mythos" – Anthropic has not released Mythos to the public, so such offers are at best a different model under a borrowed name.
FAQ
What is Claude Mythos? A series of Anthropic large language models, first disclosed on April 7, 2026, kept under restricted access because of its ability to find software vulnerabilities.
Is Claude Mythos available to the public? No. The public gets Claude Fable, the safeguarded version of the same model. Mythos requires approval (Project Glasswing or the Cyber Verification Program).
What is the latest version of Claude Mythos? Claude Mythos 5.1, released September 1, 2026 alongside Claude Fable 5.1.
What is Project Glasswing? Anthropic's program that gives selected companies and organizations access to Mythos to find and fix security vulnerabilities in critical software.
How many parameters does Claude Mythos have? Anthropic hasn't said. Industry estimates reported by the Financial Times put it at about 8 trillion.
Is Claude Mythos better than Claude Opus? For cybersecurity tasks Anthropic positions Mythos as its strongest model; the UK AI Security Institute's cyber-range test ranked Mythos first. For everyday use, the newer Opus 5.5 is comparable to Mythos 5.1 in biology and security skills, according to Anthropic.