Anthropic Releases Two Claude Models in 2026: A Public Version With Cybersecurity Limits and a Restricted One for Vetted Partners

Anthropic released two new AI models on June 9, 2026 — Claude Fable 5 and Claude Mythos 5 — with access to each determined by who is asking. Mythos 5 remains restricted to a limited set of industry partners, while Fable 5 is available to the public but with built-in limitations on what it can discuss.

Both models share the same underlying architecture, but Claude Fable 5 launches with guardrails that block responses to many queries related to cybersecurity, biology, and chemistry. Those requests are rerouted to an older model, Claude Opus 4.8. Anthropic applies the same rerouting if it suspects a user is attempting distillation — the practice of training a smaller AI model using the outputs of a larger one.

Claude Mythos 5 is being offered to partners in Project Glasswing, a consortium Anthropic established to give select organizations early access to the model’s capabilities. Anthropic said it is also providing access to “select biology researchers” and is collaborating with the US government on the rollout. The company noted it is providing unrestricted versions to these groups “until our trusted access program is available,” suggesting plans to broaden access further.

The tiered release strategy stems from concerns about the model’s ability to design hacking tools capable of finding and exploiting vulnerabilities in both new and legacy software. Anthropic first released Mythos Preview in April 2026 to a limited set of tech industry partners, giving them a head start in securing their own systems ahead of any wider availability.

Diane Penn, Anthropic’s head of product management, told WIRED that the current guardrails are intentionally cautious, meaning some benign queries may still be rerouted to the older model. She said Anthropic plans to refine its classifiers over time, but that this approach was the only safe path to a broad release at this stage.

Anthropic acknowledged in a Project Glasswing update last week that it has yet to develop the “highly robust safeguards” needed to release Mythos-level capabilities to the general public, and noted that, to its knowledge, no other AI developer has either.

Source: WIRED

This article was generated by AI and cites original sources.
Scroll to Top