CyberNotableCorroboratingDeveloping
9.4
AI security guardrails exhibit uneven efficacy across non-English languages
Dark Reading·EU · US·3 days ago
Anthropic's Opus 5 model demonstrates competitive performance in identifying software vulnerabilities compared to the Mythos 5 model. However, the model remains restricted from performing automated penetration testing or generating functional exploit code due to safety guardrails.