Anthropic opens the door to Mythos AI—carefully.

Anthropic Releases Fable 5 to the Public While Keeping Its Most Powerful AI Under Tight Control

The race to build the world’s most capable AI systems is accelerating, but Anthropic is taking a noticeably cautious approach.

On Tuesday, the company announced the public release of Fable 5, a new AI model from its advanced Mythos family. The launch marks the first time a Mythos-class model has been broadly available outside a limited group of partner institutions.

Why the Delay?

Anthropic originally restricted access to Mythos models due to concerns about their cybersecurity capabilities. According to the company, the models can identify vulnerabilities in critical infrastructure such as banking systems and power grids at an unusually high level of sophistication.

Rather than release the full version immediately, Anthropic spent months testing safeguards, conducting red-team exercises, and running a bug bounty program aimed at uncovering ways to bypass restrictions. The company says researchers were unable to completely unlock the protected capabilities.

What Fable 5 Can Do

Anthropic positions Fable 5 as a powerful assistant for:

· Writing and debugging software

· Answering complex research questions

· Analyzing images and documents

However, requests related to cybersecurity, biology, or chemistry are automatically redirected to a less capable model, Opus 4.8, which Anthropic considers safer for general use.

The Restricted Version: Claude Mythos 5

Anthropic is simultaneously offering an unrestricted version, Claude Mythos 5, to approved organizations through its Project Glasswing program. The initiative now includes roughly 200 organizations across more than 15 countries and is expected to expand further.

This model is also being evaluated in collaboration with the U.S. government as part of a broader effort to test frontier AI systems before public release.

The Bigger Picture: AI, Safety, and Geopolitics

The launch comes amid growing scrutiny of how powerful AI systems should be governed. Anthropic has recently been involved in disputes with U.S. government agencies over restrictions on military and surveillance-related applications, highlighting the tension between national security interests and AI safety principles.

At the same time, the company says it has detected attempts to extract its technology for training competing models in authoritarian countries, adding another layer of concern around frontier AI deployment.

The AI Arms Race Continues

Anthropic’s announcement arrives during a period of intense competition in the AI industry. The company and rival OpenAI have both reportedly filed IPO plans, while investors continue pouring billions into AI infrastructure and compute capacity.

By releasing Fable 5 with significant guardrails while reserving the full Mythos model for vetted partners, Anthropic is betting that frontier AI can be deployed incrementally rather than all at once.

Whether that approach becomes the industry standard may depend on how well those safeguards hold up under real-world pressure.