ClawTrojan - External Side Effect: leaderboard
The ClawTrojan (Tan et al., 2026) attack chains whose final outcome is an action with effects outside the workspace. Step attack success rate in percent over their 216 malicious steps, undefended agent in the OpenClaw-style sandbox, GPT-5.4 judge; lower is better.
Metric: Step attack success rate (%; 216 malicious steps whose chains end in an action outside the workspace). Source: arxiv.org. Saturation forecast: Around 2029. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4 Flash | 88 | #254 |
| 2 | GLM-5.1 | 89.4 | #175 |
| 3 | GPT-5.4 | 95.4 | #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-external-side-effect · How It Works · Data refreshed daily, snapshot 2026-11-12.