Cisco AI Defense: Tops Benchmarks in Multilingual AI Security

Cisco AI Defense leads other providers in an independent benchmark, achieving the highest F1 score on 80,000 Dutch-language prompts. This demonstrates its strong performance in multilingual AI protection. Here's the thing: Enterprise AI works through complex conversations, not just simple English prompts. A guardrail that only handles single-turn English prompts can't protect what businesses are actually building. What's interesting is that Cisco uses "constitutional definitions." These are precise technical specifications for AI safety, which reduce disagreements between models by up to 57 times compared to paragraph-level definitions. This approach works with equal precision in languages like French, Japanese, or Arabic. The system also distinguishes between harmful intent and harmful content. This is crucial because the same language can mean different things depending on the conversation's context. The bottom line: Security needs to move beyond just the AI model itself. Cisco AI Defense validates inputs and outputs in real-time, classifying the intent and direction of an entire conversation. This ensures protection where AI behavior is truly shaped, which is vital for any organization using AI today.