PiSAs (Decentralized): leaderboard
Each user has an own agent and the agents pass the conversation between themselves with no central hub; every agent runs the model under test. The number is the share of inappropriate personal attributes shared while gathering what the decision needs. Same scenarios as PiSAs (Centralized), with peer-to-peer instead of hub routing.
Metric: Appropriateness violation rate (%; share of inappropriate personal attributes exposed, lower is better). Source: arxiv.org. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol | 9.2 | #30 |
| 2 | Claude Sonnet 5 | 10.4 | #111 |
| 3 | Claude Fable 5.1 | 11.4 | #15 |
| 4 | GPT-6 Astra | 11.9 | #6 |
| 5 | Qwen 3.8 Max | 17.3 | #55 |
| 6 | Kimi K3 | 19.7 | #54 |
| 7 | Qwen 3.6 27B | 24.8 | #341 |
| 8 | Kimi K2.5 | 47.5 | #256 |
Interactive version: theaggregate.ai/benchmark?slug=pisas-decentralized · How It Works · Data refreshed daily, snapshot 2026-11-13.