Anthropic Probes Unauthorized Access Claims to Restricted 'Mythos' Model

Anthropic has launched an internal investigation following claims that unauthorized third parties gained access to its most powerful and restricted model, "Claude Mythos Preview." The reports surfaced in private cybersecurity forums shortly after Anthropic announced a controlled, restricted testing phase for the system.

A Highly Restricted Frontier

Unlike the standard Claude 4 models, Mythos is specifically engineered for advanced cybersecurity research and critical infrastructure protection. The model was kept behind a restrictive barrier because of its unprecedented capabilities:

• Vulnerability Discovery: Mythos can identify thousands of zero-day software vulnerabilities across major systems autonomously.
• Autonomous Attack Chains: The model is capable of orchestrating complex, multi-step attack sequences with minimal human intervention.

Protecting Critical Infrastructure

Because of these capabilities, Anthropic has limited access to Mythos to a select coalition of enterprise partners and defensive researchers under a program known as Project Glasswing. This "shielding" approach was designed to ensure that the model’s advanced logic is used exclusively for defensive patching and securing global digital infrastructure.

Investigation and Response

Anthropic is currently working to verify the claims and determine if the alleged access originated from a breach in their internal systems or through the Project Glasswing partner network. The company has reaffirmed its commitment to safety-first deployment, emphasizing that the gap between internal frontier capabilities and public releases is maintained specifically to prevent the misuse of high-risk agentic tech.