Benchmark RadarAI BENCHMARK PROFILE
AICompanionBench
AICompanionBench provides a dataset of 2,123 human-AI companion conversations with safety risk annotations to evaluate LLMs-as-judges for detecting unsafe interactions.
- Released
- 2026-06-03
- Readiness
- Runnable
- Primary field
- Robotics & Autonomous Systems
Why it matters
Supports the development of automated safety monitoring for AI companion platforms by providing a fine-grained benchmark for risk detection.
Motivation
As AI companion platforms such as Replika and Character.AI rapidly grow, concerns about unsafe human-AI interactions have intensified.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.