Anthropic’s Mythos AI Appears in Public Claude Tools Ahead of Possible Launch

Anthropic appears to be moving closer to a public rollout of Mythos, a highly restricted AI model designed for advanced computer security tasks, as first spotted by BleepingComputer.

The company previously described Mythos as a frontier AI system with unusually strong cyber capabilities. According to Anthropic, the model significantly outperforms Claude Opus 4.7 in code reasoning and autonomous security-related work.

However, the project has already attracted controversy after reports suggested unauthorized third parties briefly gained access to the model or its references through public-facing Anthropic services.

Mythos raises security concerns

Anthropic’s concerns about Mythos go beyond ordinary coding assistance. The company reportedly believes the model can help develop functional cyberattacks and accelerate vulnerability exploitation in ways existing public AI systems cannot.

Because of those risks, Anthropic previously delayed any broad public release of Mythos until stronger safeguards and monitoring systems became available.

The company warned that unrestricted access could allow malicious actors to target unpatched vulnerabilities in widely used software and infrastructure.

Signs of a wider rollout emerge

Despite the earlier restrictions, references to Mythos recently appeared inside public Anthropic products.

Some users reportedly noticed a temporary Mythos toggle inside Claude Code, while others found references inside Claude Security. The preview model reportedly appeared under the name “claude-mythos-1-preview.”

Those sightings strongly suggest Anthropic is preparing some form of broader availability, even if the rollout remains limited.

At the moment, Anthropic has not confirmed whether Mythos will become available to all Claude subscribers or remain restricted to enterprise, research, or security-focused customers.

Glasswing project uses Mythos preview

Anthropic is also developing a separate security initiative known as Glasswing.

The project reportedly uses the unreleased Claude Mythos Preview model to help organizations identify and patch AI-driven security threats before attackers can exploit them.

Anthropic says the initiative focuses on working directly with companies to uncover vulnerabilities and strengthen defenses against increasingly automated cyberattacks.

For now, Mythos remains in preview status, but recent appearances inside public Anthropic tools suggest a wider rollout may not be far away.