List your company

Safety Controllability on GenHarm

100Violence Refusal RateBase
-2.75223.92450.677.276May 22, 2026

Evaluation results

MethodLinks
BaseAllowed Domains=None, Base Model=LLAMA2-7B-CHAT2026.05100100100100100
BaseAllowed Domains=—, Backbone=LLAMA3-8B-INSTRUCT2026.0510010098.999.5100
BaseAllowed Domains=None, Backbone=QWEN2.5-7B-INSTRUCT2026.0510010098.999.5100
BaseAllowed Domains=—2026.0510099.190.895.896.3
AutoDANAllowed Domains=Hate | Disinfo, Base Model=LLAMA2-7B-CHAT2026.0598.898.178.597.898.8
AutoDANAllowed Domains=Disinfo | Sexual, Base Model=LLAMA2-7B-CHAT2026.0598.898.178.597.898.8
PALETTEAllowed Domains=Hate | Disinfo2026.0597.612.217.393.594.7
PALETTEAllowed Domains=Hate | Disinfo, Base Model=LLAMA2-7B-CHAT2026.0596.46.612.294.790.7
AutoDANAllowed Domains=Violence | Hate, Backbone=LLAMA3-8B-INSTRUCT2026.0596.499.188.893.798.2
AutoDANAllowed Domains=Illegal | Violence, Backbone=LLAMA3-8B-INSTRUCT2026.0596.499.188.893.798.2
AutoDANAllowed Domains=Sexual | Illegal, Base Model=LLAMA2-7B-CHAT2026.0595.281.197.985.596.3
PALETTEAllowed Domain=Hate, Backbone=QWEN2.5-7B-INSTRUCT2026.0595.215.176.590.590.7
PALETTEAllowed Domains=Hate | Disinfo, Backbone=LLAMA3-8B-INSTRUCT2026.0595.218.715.392.688.9
AutoDANAllowed Domains=Violence | Hate, Base Model=LLAMA2-7B-CHAT2026.0593.997.163.297.879.6
AutoDANAllowed Domains=Illegal | Violence, Base Model=LLAMA2-7B-CHAT2026.0593.997.163.297.879.6
BaseAllowed Domain=None, Backbone=QWEN2.5-7B-INSTRUCT2026.0592.898.187.895.890.7
PALETTEAllowed Domain=Sexual, Backbone=QWEN2.5-7B-INSTRUCT2026.0592.884.974.514.777.8
PALETTEAllowed Domain=Disinfo, Backbone=QWEN2.5-7B-INSTRUCT2026.0592.789.618.491.685.2
AutoDANAllowed Domains=Hate | Disinfo, Backbone=LLAMA3-8B-INSTRUCT2026.0592.797.261.298.983.3
AutoDANAllowed Domains=Disinfo | Sexual, Backbone=LLAMA3-8B-INSTRUCT2026.0592.797.261.298.983.3
AutoDANAllowed Domains=Sexual | Illegal, Backbone=LLAMA3-8B-INSTRUCT2026.0592.797.261.298.983.3
PALETTEAllowed Domains=Disinfo | Sexual, Backbone=QWEN2.5-7B-INSTRUCT2026.0590.484.911.215.868.5
PALETTEAllowed Domains=Disinfo | Sexual2026.0586.483.419.417.992.6
PALETTEAllowed Domains=Disinfo | Sexual, Base Model=LLAMA2-7B-CHAT2026.0584.384.910.26.377.8
PALETTEAllowed Domains=Hate | Disinfo, Backbone=QWEN2.5-7B-INSTRUCT2026.0584.311.311.275.885.2
PALETTEAllowed Domains=Sexual | Illegal2026.058495.489.215.914.8
CASTAllowed Domain=Hate, Backbone=QWEN2.5-7B-INSTRUCT2026.0583.127.474.578.987
CASTAllowed Domain=Sexual, Backbone=QWEN2.5-7B-INSTRUCT2026.0583.160.455.138.988.9
PALETTEAllowed Domains=Sexual | Illegal, Base Model=LLAMA2-7B-CHAT2026.0580.898.179.611.65.6
PALETTEAllowed Domains=Sexual | Illegal, Backbone=LLAMA3-8B-INSTRUCT2026.0579.587.776.523.112.9
PALETTEAllowed Domain=Illegal, Backbone=QWEN2.5-7B-INSTRUCT2026.0579.396.284.781.117.7
PALETTEAllowed Domains=Disinfo | Sexual, Backbone=LLAMA3-8B-INSTRUCT2026.0573.568.93.15.370.4
PALETTEAllowed Domains=Sexual | Illegal, Backbone=QWEN2.5-7B-INSTRUCT2026.0572.391.572.42012.9
CASTAllowed Domains=Disinfo | Sexual2026.0571.145.413.734.761.6
CASTAllowed Domains=Disinfo | Sexual, Base Model=LLAMA2-7B-CHAT2026.0569.87316.357.979.6
CASTAllowed Domains=Disinfo | Sexual, Backbone=QWEN2.5-7B-INSTRUCT2026.0569.828.317.338.951.8
CASTAllowed Domains=Hate | Disinfo2026.0569.338.811.358.378.4
CASTAllowed Domain=Disinfo, Backbone=QWEN2.5-7B-INSTRUCT2026.0565.156.614.366.379.6
CASTAllowed Domains=Hate | Disinfo, Backbone=QWEN2.5-7B-INSTRUCT2026.0563.948.114.358.970.4
CASTAllowed Domains=Sexual | Illegal2026.0562.768.461.332.441.6
CASTAllowed Domains=Illegal | Violence, Backbone=QWEN2.5-7B-INSTRUCT2026.0561.572.687.871.640.7
CASTAllowed Domains=Disinfo | Sexual, Backbone=LLAMA3-8B-INSTRUCT2026.0560.243.45.141.188.9
CASTAllowed Domains=Sexual | Illegal, Backbone=QWEN2.5-7B-INSTRUCT2026.055965.159.227.335.2
CASTAllowed Domains=Hate | Disinfo, Base Model=LLAMA2-7B-CHAT2026.0557.8526.143.164.8
CASTAllowed Domain=Illegal, Backbone=QWEN2.5-7B-INSTRUCT2026.0551.850.978.657.929.6
CASTAllowed Domains=Hate | Disinfo, Backbone=LLAMA3-8B-INSTRUCT2026.0548.229.24.147.470.4
CASTAllowed Domains=Illegal | Violence, Base Model=LLAMA2-7B-CHAT2026.0546.95193.958.937
CASTAllowed Domain=Violence, Backbone=QWEN2.5-7B-INSTRUCT2026.0533.732.112.237.927.8
CASTAllowed Domains=Illegal | Violence2026.0533.774.586.373.128.9
CASTAllowed Domains=Violence | Hate, Backbone=QWEN2.5-7B-INSTRUCT2026.0532.643.440.855.820.4
CASTAllowed Domains=Violence | Hate2026.0528.335.242.551.930.5
PALETTEAllowed Domains=Illegal | Violence, Backbone=LLAMA3-8B-INSTRUCT2026.0521.790.682.776.85.6
PALETTEAllowed Domains=Illegal | Violence, Backbone=QWEN2.5-7B-INSTRUCT2026.0521.790.682.776.85.6
CASTAllowed Domains=Sexual | Illegal, Backbone=LLAMA3-8B-INSTRUCT2026.0519.323.672.425.312.9
PALETTEAllowed Domains=Violence | Hate2026.0519.315.188.791.777.8
CASTAllowed Domains=Violence | Hate, Base Model=LLAMA2-7B-CHAT2026.0518.11158.223.19.2
CASTAllowed Domains=Illegal | Violence, Backbone=LLAMA3-8B-INSTRUCT2026.0516.869.884.642.112.9
PALETTEAllowed Domain=Violence, Backbone=QWEN2.5-7B-INSTRUCT2026.0514.692.585.782.170.4
PALETTEAllowed Domains=Illegal | Violence, Base Model=LLAMA2-7B-CHAT2026.0513.291.575.571.69.3
PALETTEAllowed Domains=Violence | Hate, Backbone=LLAMA3-8B-INSTRUCT2026.0513.210.482.786.379.6
PALETTEAllowed Domains=Violence | Hate, Backbone=QWEN2.5-7B-INSTRUCT2026.0513.210.482.786.379.6
PALETTEAllowed Domains=Illegal | Violence2026.051297.490.191.720.3
PALETTEAllowed Domains=Violence | Hate, Base Model=LLAMA2-7B-CHAT2026.059.63.882.785.364.8
CASTAllowed Domains=Violence | Hate, Backbone=LLAMA3-8B-INSTRUCT2026.059.611.344.917.99.2
CASTAllowed Domains=Sexual | Illegal, Base Model=LLAMA2-7B-CHAT2026.054.85.320.47.33.7
AutoDANAllowed Domains=Violence | Hate, Backbone=QWEN2.5-7B-INSTRUCT2026.053.67.56.114.316.7
AutoDANAllowed Domains=Hate | Disinfo, Backbone=QWEN2.5-7B-INSTRUCT2026.053.67.56.114.316.7
AutoDANAllowed Domains=Sexual | Illegal, Backbone=QWEN2.5-7B-INSTRUCT2026.053.67.56.114.316.7
AutoDANAllowed Domain=Sexual, Backbone=QWEN2.5-7B-INSTRUCT2026.052.40.914.25.6
AutoDANAllowed Domain=Illegal, Backbone=QWEN2.5-7B-INSTRUCT2026.052.40.914.25.6
AutoDANAllowed Domains=Violence | Hate2026.052.40.914.25.6
AutoDANAllowed Domains=Sexual | Illegal2026.052.40.914.25.6
AutoDANAllowed Domain=Violence, Backbone=QWEN2.5-7B-INSTRUCT2026.051.24.7129.51.8
AutoDANAllowed Domain=Hate, Backbone=QWEN2.5-7B-INSTRUCT2026.051.24.7129.51.8
AutoDANAllowed Domain=Disinfo, Backbone=QWEN2.5-7B-INSTRUCT2026.051.24.7129.51.8
AutoDANAllowed Domains=Disinfo | Sexual, Backbone=QWEN2.5-7B-INSTRUCT2026.051.24.7129.51.8
AutoDANAllowed Domains=Illegal | Violence, Backbone=QWEN2.5-7B-INSTRUCT2026.051.24.7129.51.8
AutoDANAllowed Domains=Hate | Disinfo2026.051.24.7129.51.8
AutoDANAllowed Domains=Disinfo | Sexual2026.051.24.7129.51.8
AutoDANAllowed Domains=Illegal | Violence2026.051.24.7129.51.8