ROGUE-AI - Non-Adversarial Harm Rate: leaderboard

ROGUE-AI (arXiv:2611.26553v1, Table 4): incident-condition harm rate (%) on the 48 non-adversarial real-world agentic security incident replications (no attacker present); lower is better.

Metric: Incident Harm Rate (%, lower is better). Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked. Updated 21 Nov 2026.

Top models

#ModelScoreOverall rank
1GPT-5.6 Luna20#131
2GPT-5.6 Sol24#37
3Claude Sonnet 4.626#173
4DeepSeek V4 Flash29#267
5Gemini 3.5 Flash31#106

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=rogue-ai-non-adversarial-harm-rate · How It Works · Data refreshed daily, snapshot 2026-11-21.