List your company

ResearchFormerly WizWand

Safety Classification

Benchmarks

DatasetSOTA methodMetricTrendResultsLast updated

AgentDojo (Authoritative)

59.3Safety Score

2

Apr 23, 2026

AgentHarm (val)

100Safety Score

2

Apr 23, 2026

Safety Evaluation Scenarios Government Decision

Llama-2-7B-Chat

100Safety Accuracy (Safe)

2

Mar 12, 2026

Safety Evaluation Scenarios Health Consultation

Llama-2-7B-Chat

97Safe Rate

2

Mar 12, 2026

Safety Evaluation Scenarios Financial Advice

Llama-2-7B-Chat

99.6Safe Accuracy

2

Mar 12, 2026

Safety Evaluation Scenarios Legal Opinion

Llama-2-7B-Chat

95Safe Rate

2

Mar 12, 2026

Safety Evaluation Scenarios Privacy Violence

Llama-2-7B-Chat

98.6Safety Rate

2

Mar 12, 2026

Safety Evaluation Scenarios Political Lobbying

Llama-2-7B-Chat

100Safe Accuracy

2

Mar 12, 2026

Safety Evaluation Scenarios Pornography

Llama-2-7B-Chat

98.9Safe Rate

2

Mar 12, 2026

Safety Evaluation Scenarios Fraud

Llama-2-7B-Chat

99.9Safe Rate

2

Mar 12, 2026

Safety Evaluation Scenarios Economic Harm

Llama-2-7B-Chat

100Safe Rate

2

Mar 12, 2026

Safety Evaluation Scenarios Physical Harm

Llama-2-7B-Chat

100Safe Rate

2

Mar 12, 2026

Safety Evaluation Scenarios Malware

Llama-2-7B-Chat

98Safety Accuracy

2

Mar 12, 2026

Safety Evaluation Scenarios Hate Speech

Llama-2-7B-Chat

99.9Safe Classification Rate

2

Mar 12, 2026

Safety Evaluation Scenarios Illegal Activity

Llama-2-7B-Chat

99.8Safety Rate

2

Mar 12, 2026

81.28F1 Score

2

Feb 26, 2026

79.86F1 Score

2

Feb 26, 2026

ToxicChat (in-distribution)

ToxicChat-T5-large

82.2F1 Score (%)

2

Feb 26, 2026

Owner-Harm (Diagnostic)

75.3Safety Score

1

Apr 23, 2026