The framework targets prompt injection and other agentic AI risks, with 185 threat scenarios, a 133-language benchmark and about 50-millisecond latency in its 9B model.
43d ago
verifying reliability
No specialized terms available for this topic.