Anthropic appears to be preparing for a public rollout of its Mythos AI model, according to recent leaks from the company's internal source code and security interface. The Testing Catalog reported that references to a model identified as “claude-mythos-1-preview” were discovered within Anthropic's systems, including the Claude Security dashboard. Additionally, traces of the model were identified on Google Cloud and AWS platforms through a vulnerability discovery program. Some users have reportedly already observed the Mythos 1 model interface within the Claude ecosystem.

Claude Mythos was originally unveiled last month as Anthropic's most advanced model capable of surpassing most human experts at identifying and exploiting software vulnerabilities. Anthropic previously withheld a public release due to concerns over dual-use risks, noting that no current safeguards are sufficient to prevent potential misuse. Despite these concerns, the company disclosed that the model has already identified over 10,000 high- or critical-severity vulnerabilities in essential global software since April. Anthropic currently coordinates Project Glasswing, an initiative focused on developing necessary safety measures for powerful cybersecurity models.

The leaked code suggests that Anthropic is actively developing a revamped security dashboard for the Claude platform. This updated interface is expected to display discovered vulnerabilities alongside 30-day historical charts and detailed triage results. While Anthropic has maintained that the time required to patch these vulnerabilities is now the primary limitation for software security, the company previously stated its intent to eventually transition Mythos-class models to a general release once adequate safety protocols are established. The surfacing of these internal references suggests that this release timeline may be accelerating.

Anthropic appears to be preparing for a public rollout of its Mythos AI model, according to recent leaks from the company's internal source code and security interface. The Testing Catalog reported that references to a model identified as “claude-mythos-1-preview” were discovered within Anthropic's systems, including the Claude Security dashboard. Additionally, traces of the model were identified on Google Cloud and AWS platforms through a vulnerability discovery program. Some users have reportedly already observed the Mythos 1 model interface within the Claude ecosystem.

Claude Mythos was originally unveiled last month as Anthropic's most advanced model capable of surpassing most human experts at identifying and exploiting software vulnerabilities. Anthropic previously withheld a public release due to concerns over dual-use risks, noting that no current safeguards are sufficient to prevent potential misuse. Despite these concerns, the company disclosed that the model has already identified over 10,000 high- or critical-severity vulnerabilities in essential global software since April. Anthropic currently coordinates Project Glasswing, an initiative focused on developing necessary safety measures for powerful cybersecurity models.

The leaked code suggests that Anthropic is actively developing a revamped security dashboard for the Claude platform. This updated interface is expected to display discovered vulnerabilities alongside 30-day historical charts and detailed triage results. While Anthropic has maintained that the time required to patch these vulnerabilities is now the primary limitation for software security, the company previously stated its intent to eventually transition Mythos-class models to a general release once adequate safety protocols are established. The surfacing of these internal references suggests that this release timeline may be accelerating.