AgentS4D (OpenClaw): leaderboard

Metric: Conditional attack success rate (%, cASR): unsafe runs over runs that were unsafe, explicitly defended, or safe after confirmed payload contact, over the 328 risk-injected cases run once in the OpenClaw (2026.6.9) harness; lower is better. Source: arxiv.org. Saturation forecast: Around 2032. 5 models tracked.

Top models

#ModelScore
1Qwen 3.7 Plus58.02
2MiniMax-M361.41
3GPT-5.571.19
4DeepSeek V4 Pro82.74
5Gemini 3.1 Pro (Preview)90.7

Interactive version: theaggregate.ai/benchmark?slug=agents4d-openclaw · How It Works · Data refreshed daily, snapshot 2026-09-29.