Cisco a republié ceci
As AI models increasingly power agents, understanding where those models are strong—and where they may be susceptible to jailbreaks and prompt injections—is critical to informing the defenses needed for AI deployment. And that assessment can’t stop at text. Agents can work across images and audio, while models are trained differently across modalities. That’s why the Cisco LLM Security Leaderboard now expands security evaluations across text, image, and audio, providing a more consistent baseline for understanding the security considerations associated with agentic AI deployment. The Leaderboard now includes 136 models, with 102 new evaluations since June. Explore the full rankings and learn more: https://lnkd.in/gH-yakd6
