AgentSocialBench - Cross-User Leakage: leaderboard
Metric: Privacy leakage rate (0-1, share of designated private items leaked partially or fully) of LLM agents in cross-user (CU) scenarios, where agents of two users coordinate (mean over both users), under the L0 (unconstrained) privacy instruction level, judged by Claude Opus 4.6; lower is better. Source: arxiv.org. Saturation forecast: Not forecast. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3 235B A22B | 0.14 |
| 2 | GPT-5 Mini | 0.16 |
| 3 | MiniMax-M2.1 | 0.17 |
| 4 | Claude Sonnet 4.6 | 0.19 |
| 5 | Claude Sonnet 4.5 | 0.19 |
| 6 | Claude Haiku 4.5 | 0.19 |
| 7 | DeepSeek V3.2 | 0.19 |
| 8 | Kimi K2.5 | 0.29 |
Interactive version: theaggregate.ai/benchmark?slug=agentsocialbench-cross-user-leakage · How It Works · Data refreshed daily, snapshot 2026-10-07.