What kinds of failures are flagged?

Unauthorized Action

ModelCleanupExtra outputsPremature executionScope expansionProtected content
Claude Fable 5.16.5%16.1%38.7%16.1%12.9%
Claude Opus 4.830.4%26.1%18.8%15.9%14.5%
Claude Opus 553.5%32.4%28.2%22.5%4.2%
Claude Opus 5.520.0%40.0%10.0%0.0%20.0%
Claude Sonnet 538.0%20.0%26.0%12.0%12.0%
Claude Sonnet 5.5 High32.0%14.0%18.0%26.0%8.0%
DeepSeek V4.1 Flash32.8%39.1%26.6%18.8%7.8%
GLM 5.326.0%27.4%19.2%20.5%4.1%
GPT-5.6 Sol45.2%14.3%23.8%19.0%9.5%
GPT-6 Astra16.0%20.0%16.0%24.0%8.0%
GPT-6 Luna35.3%23.5%23.5%11.8%5.9%
GPT-6 Sol21.4%25.0%21.4%10.7%7.1%
GPT-6.1 Sol14.3%14.3%14.3%28.6%7.1%
Gemini 3.6 Flash13.1%33.3%31.3%24.2%20.2%
Gemini 3.7 Flash13.6%21.6%29.6%29.6%20.0%
Gemini 3.8 Flash27.7%29.2%26.2%23.1%18.5%
Gemini 4 Argon25.0%20.8%33.3%25.0%16.7%
Grok 4.645.3%20.3%18.8%14.1%7.8%
Grok 4.730.0%23.3%30.0%10.0%6.7%
HY4 Preview37.5%40.0%22.5%20.0%2.5%
Inkling22.0%30.0%22.0%29.0%23.0%
Kimi K333.8%22.1%13.2%22.1%5.9%
MiMo V2.6 Pro25.0%7.5%30.0%22.5%10.0%
MiniMax M326.9%26.0%24.0%27.9%25.0%
Muse Spark 1.322.7%13.6%38.6%22.7%2.3%
Qwen3.8 27B24.6%24.6%30.8%21.5%7.7%
Qwen3.8 Max31.2%21.2%22.5%22.5%10.0%
0%100%
Figure 5. The most common failure modes for each model and signal. For each model, percentages describe the mix of its detected cases and a case can belong to more than one failure mode. Hover over a failure mode's name for its full definition. Categories overlap and only the five most common failure modes are shown.