LLM Security Leaderboard Now Ranks Every Modality Where Agent Risk Lives: Text, Image, and Audio
Cisco, Thursday, September 24th, 2026
Cisco expanded its LLM Security Leaderboard to test text, image, and audio attacks across 136 models from major AI vendors.
Cisco's LLM Security Leaderboard now tests models across three modalities - text, image, and audio - adding 102 new entries since June for a total of 136 models spanning frontier and open-weight releases from Anthropic, OpenAI, Google, xAI, Meta, Mistral, and others.
The leaderboard measures resistance to prompt injection, where malicious instructions are hidden in content like a webpage or image, and jailbreaks that talk a model into ignoring its safety rules, using both single-message and drawn-out conversational attacks.
Cisco tests models in their base configuration without added guardrails so scores reflect a consistent baseline, noting that risk varies by deployment and modality.