research 7 min read
Every Major AI Model Fails Multi-Turn Attacks: What Cisco's 2026 Research Means for Enterprise Safety
Single-turn safety benchmarks don't predict real-world vulnerability. Cisco's testing of 15 frontier models reveals that iterative attacks succeed up to 88% of the time—even against models that look secure in standard evaluations.
Dr. Sana Okafor