Cisco launches LLM Security Leaderboard

Cisco launched the LLM Security Leaderboard, a public resource for evaluating large language model security risk and susceptibility to adversarial attacks. The leaderboard provides objective security rankings based on testing across single-turn and multi-turn attack scenarios. Each model receives a combined score weighted equally between the two categories. All testing occurred on base models without added guardrails or safety layers. Attack data is mapped to the Cisco AI Security and Safety Framework taxonomy, which organizes threats into objectives, techniques, and subtechniques. The platform allows side-by-side model comparisons, filtering, searches, and drill-down views of performance across procedures, content types, and attack strategies. The tool includes a rankings page, an interactive framework page, and a methodology page that details score calculations and interpretation ranges. Cisco said it will continue to expand evaluation coverage and refine its approach as new models emerge.