Cisco Expands LLM Security Leaderboard to Text, Image and Audio, Ranking 136 Models
aminkarbasi · x · 2026-09-25
Cisco has updated its LLM Security Leaderboard to evaluate model security across text, image, and audio modalities—the input surfaces where agents are most exposed—adding 102 new evaluations since June and now covering 136 models.
The leaderboard focuses on two attack types:
- Prompt injection: malicious instructions hidden in content the model processes, such as webpages or images
- Jailbreaks: talking a model into ignoring its own safety rules
Cisco notes that risk varies by deployment: a model wired into a browsing agent faces a very different exposure profile than one in a chat window, so knowing where a model is weak matters as much as where it is strong.
More from Safety
- Google now estimates Canadian users' age from search and YouTube history — WellsLucasSanto · 2026-09-25
- Court hears AI-generated songs man wrote for his mistress as evidence in wife-murder trial — Polymarket · 2026-09-25
- Altimeter CEO: OpenAI's Navier–Stokes-solving model withheld amid safety and govt scrutiny — rohanpaul_ai · 2026-09-25
- Palo Alto bundles four products into 'Secure AI Coding' platform for AI coding agents — shashib · 2026-09-25
- Nathan Calvin presses Anthropic: when would you unilaterally halt AI development? — dgrobinson · 2026-09-25
- 10 House co-sponsors sign on to bill banning artificial superintelligence — erikphoel · 2026-09-25