AICompanionBench: Benchmarking LLMs-as-Judges for AI Companion Safety

arXiv:2606.04867Submitted Jun 4, 20260 benchmark results

Authors pending

Abstract

First public benchmark (2,123 real Replika conversations) for fine-grained safety risk categories; reveals LLMs struggle with implicit unsafe interactions like manipulation.

Tasks

edit

• Benchmarks

Results

No benchmark results recorded yet.

submit

Benchmark results referencing this paper haven't been added to the registry yet. If you have a reproduction, submit it →

CodeSOTA extraction

Benchmark evidence

edit

AICompanionBench: per-category accuracy of LLM-as-judge on 9 safety risk categories (e.g., manipulation, self-harm)

Add or update benchmark results

Logged-in editor · benchmark trail

Three places to go from here.

Index

All papers

All tracked papers in the registry, with benchmark result, model, and leaderboard linkage where available.

Replacement

Papers with Code is dead — alternatives

What replaced PWC for each use case: LLMs, OCR, speech, vision, robotics.

Top hub

Benchmarks

Every benchmark in Benchmarks.