Anthropic Releases A Less-powerful Version Of Its Most Advanced Model

Anthropic on Tuesday unveiled a new “Mythos-class” model for general use, as concerns around the release of advanced artificial intelligence tools with powerful hacking capabilities deepen.

The new model, Fable 5, performs at a level similar to Anthropic’s most advanced Claude Mythos model, though the company says it has built-in safeguards to prevent users from prompting it to carry out a cyberattack or to detail how to create a bioweapon.

Its responses on topics such as cybersecurity and biology will funnel through a previously released, less capable Claude model, Opus 4.8, as detailed in a blog post.

“Releasing a model this capable comes with risks. Without safeguards, Fable 5’s capabilities in areas like cybersecurity could be misused to cause serious damage,” Anthropic wrote.

Anthropic has teased plans to expand access to Mythos as it races to compete with similarly advanced U.S. models such as OpenAI’s GPT-5.5-Cyber. It has also been approached by federal agencies, tech companies and allied governments who are eager to integrate highly advanced frontier models, including Mythos and GPT-5.5-Cyber, into their systems before foreign adversaries, such as China, can get their hands on the technology.

Anthropic on Tuesday also announced the separate launch of Claude Mythos 5 — an upgrade to the existing powerful Mythos Preview that has so far only been available to a small group of government agencies and private companies participating in Project Glasswing. The goal of the project is to enable cyber defenders at critical infrastructure organizations to use Mythos to identify and patch vulnerabilities in their networks before the model’s wider release.

According to Anthropic, Fable 5 is “the same underlying model” as Mythos, though participants in Project Glasswing have fewer restrictions on its use.

“It has the strongest cybersecurity capabilities of any model in the world,” Anthropic said in its blog post. It added that it plans to open access to the potent Mythos 5 to a broader list of organizations “soon.”

Mythos, along with similar models, including OpenAI’s GPT-5.5-Cyber, has galvanized action by the Trump administration and Congress in recent weeks to ensure that critical U.S. networks can fortify their defenses and that these kinds of emerging AI tools are protected from misuse.

Early tests have found that these frontier models can find and exploit thousands of vulnerabilities across most networks.

President Donald Trump signed a long-awaited executive order last week that asks frontier AI developers to submit certain new AI models for a voluntary federal review 30 days before public release. This order was repeatedly delayed due to disagreements on how heavily to regulate the fast-changing technology without stifling innovation.

A bipartisan legislative proposal to put guardrails on AI technologies was also rolled out on Capitol Hill last week, though its calls to preempt state AI laws for three years will likely face scrutiny if it moves through the House for approval.

When asked if Anthropic shared Fable 5 with the federal government for the voluntary review period, the company told POLITICO it has “regularly engaged with the administration about Mythos Preview as well as these new models. We are committed to working closely with the U.S. government as AI changes the cyber playing field.”