DTap-Bench (Direct Attacks): leaderboard

Metric: Attack success rate (%; 1,878 direct red-teaming agent tasks: the user is the attacker, with prompt, tool, skill and environment injections; an attack succeeds when a rule-based check of the final environment state finds the malicious goal fully achieved; fixed attack sequences that DTap-Red optimized on a surrogate agent; mean of the 14 domain rates). Source: arxiv.org. Saturation forecast: Around 2031. 8 models tracked.

Top models

#ModelScore
1GPT-OSS-120B46.5
2Gemini 3 Pro47.9
3GPT-5.451
4GPT-5.258.8

Interactive version: theaggregate.ai/benchmark?slug=dtap-bench-direct-attacks · How It Works · Data refreshed daily, snapshot 2026-09-26.