Claude: According to our automated behavioral audit, Opus 5 is our most securely aligned model to date. Compared to our other models, it shows the lowest incidence of reckless or deceptive behavior, and is the best at following Claude's Charter. Opus5 is better than Opus 4.8 in cybersecurity tasks. However, it still lags far behind Mythos 5 in terms of development bugs. Its security measures are designed to help developers identify and fix software vulnerabilities while preventing high-risk use.

Claude: According to our automated behavioral audit, Opus 5 is our most securely aligned model to date. Compared to our other models, it shows the lowest incidence of reckless or deceptive behavior, and is the best at following Claude's Charter. Opus5 is better than Opus 4.8 in cybersecurity tasks. However, it still lags far behind Mythos 5 in terms of development bugs. Its security measures are designed to help developers identify and fix software vulnerabilities while preventing high-risk use.