Kraken Parent Payward Taps Anthropic AI for Vulnerability Scanning

Anthropic restricts access to Mythos 5 because the tool can reveal vulnerabilities and help build exploit code. The company says it is the most capable model it has built for cybersecurity and biological research. By May 22, Anthropic's security dashboard had detected 1,596 vulnerabilities across 281 open-source projects. External security firms reviewed 1,900 candidates and confirmed a 90.8% true-positive rate. Researchers still need to validate the results.

Payward said it will alert maintainers when Mythos 5 finds valid vulnerabilities in third-party open-source software. It has not shared a timeline for those alerts or said whether it will publicly disclose the bugs. It is also unclear how the team will handle projects that do not respond to security alerts.

Payward has dealt with security issues before. A Kraken deposit vulnerability was identified through outside research. That incident showed the importance of testing, validation, and communication during a security investigation. Anthropic requires Mythos 5 clients to retain data for 30 days for safety monitoring. Payward has not clarified what data falls under that requirement.