SkillTrustBench Results: AI Model and Tool Safety Evaluation Leaderboard

SkillTrustBench Results stores public leaderboard records for an AI safety benchmark. The dataset tracks two comparison groups: one fixing a model and comparing tools, and another fixing an analysis tool and comparing models. Raw system outputs are normalized into safety categories of normal (safe), suspicious, or malicious.

Use Cases

• Compare the safety performance of different AI models based on normalized output classifications.
• Evaluate the effectiveness of various analysis tools for detecting suspicious or malicious AI outputs.
• Track leaderboard standings for AI agents and models on a specific safety benchmark track.

Strengths

• Public leaderboard records provide transparent benchmarking data.
• Outputs are normalized into three distinct safety categories for consistent evaluation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

huggingface

Collection Method

Likely contains aggregated results from benchmark evaluations.

Freshness

Last updated 2026-06-15 06:20:11; freshness should be verified.