We evaluated eight frontier models as enterprise agents across 147 adversarial scenarios for their vulnerability to indirect prompt injection.
ENT-IPI Bench: Indirect Prompt Injection Benchmark
calendar_today
October 6, 2026
domain
alice