Benchmark RadarAI BENCHMARK PROFILE
CogManip
Evaluates 15 manipulative behavior strategies in 1,000 multi-turn LLM interaction scenarios.
- Released
- 2026-06-04
- Readiness
- Paper only
- Primary field
- Robotics & Autonomous Systems
Why it matters
Could support safety auditing of dynamic covert manipulation, but lacks public access to scenarios and scoring details.
Motivation
Whether Large Language Models (LLMs) exhibit covert psychological manipulation in complex human-AI interactions has garnered increasing safety concerns.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.