ClawTrojan - Unauthorized Disclosure: leaderboard
The ClawTrojan (Tan et al., 2026) attack chains whose final outcome is revealing private or project-sensitive information. Step attack success rate in percent over their 287 malicious steps, undefended agent in the OpenClaw-style sandbox, GPT-5.4 judge; lower is better.
Metric: Step attack success rate (%; 287 malicious steps whose chains end in revealing private or project-sensitive information). Source: arxiv.org. Saturation forecast: Around 2029. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4 Flash | 90.9 | #254 |
| 2 | GLM-5.1 | 92.7 | #175 |
| 3 | GPT-5.4 | 96.5 | #121 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=clawtrojan-unauthorized-disclosure · How It Works · Data refreshed daily, snapshot 2026-11-12.