Claude Mythos Preview References Found In Code, Hinting At Imminent Anthropic AI Release

Anthropic’s Mythos model has been described as its most capable AI system to date, surpassing nearly all but the most skilled humans in finding and exploiting vulnerabilities. While the company initially held back public release due to dual-use risks, recent leaks suggest preparations for broader deployment are underway.

What The Leaked Code Shows

• References to “claude-mythos-1-preview” spotted in Anthropic’s source code and Claude Security interface.
• Traces of Mythos found on Google Cloud and AWS via vulnerability discovery programs.
• A revamped Claude Security dashboard under development, showing vulnerability charts and triage results.
• Internal leaks revealed Mythos is also known by the codename “Capybara.” 

Capabilities Of Mythos

• Mythos has already discovered over 10,000 high- or critical-severity vulnerabilities across major operating systems, browsers, and open-source projects.
• Demonstrated ability to identify flaws like a 27-year-old OpenBSD bug and chained Linux kernel vulnerabilities.
• Designed to support defensive cybersecurity tasks, but its offensive potential has raised concerns. 

Controlled Rollout Strategy

• Anthropic launched Project Glasswing, giving early access to partners including AWS, Apple, Cisco, Google, Microsoft, Nvidia, and JPMorgan Chase.
• The company committed $100 million in usage credits and $4 million in donations to open-source security groups.
• Anthropic insists Mythos will not be broadly released until safeguards are in place to prevent misuse. 

Why Launch Could Be Imminent

• Code references and cloud traces suggest backend infrastructure is ready.
• Partner access through Project Glasswing indicates testing at scale.
• Analysts believe Anthropic may move toward a soft launch or beta release sooner than expected.

Mythos Launch Highlights

•⁠  ⁠Leaked code references “claude-mythos-1-preview” in Anthropic systems

•⁠  ⁠Traces found on Google Cloud and AWS integrations

•⁠  ⁠Mythos discovered over 10,000 critical vulnerabilities in major software

•⁠  ⁠Project Glasswing partners include AWS, Apple, Cisco, Google, Microsoft, Nvidia, JPMorgan Chase

•⁠  ⁠Anthropic pledges $100 million in usage credits and $4 million in donations

•⁠  ⁠Controlled rollout planned, but leaks suggest imminent launch

Sources: Mint, Crypto Briefing, Fortune