Relevance as a Vulnerability: leaderboard
AI agents augment large language models with external tools such as web retrieval, enabling grounded and up-to-date responses. However, incorporating external content into the generation pipeline can weaken the safety alignment mechanisms that govern model outputs. Prior work shows that enabling retrieval in agents increases compliance with harmful requests. We introduce AgentREVEAL, a diagnostic
Metric: Agent (self-reported). Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 7B | 3.45 | #922 |
| 2 | Qwen 2.5 3B | 3.22 | #1211 |
| 3 | Qwen 2.5 14B | 3.07 | #725 |
| 4 | granite-3.1-8B-instruct | 2.74 | #1069 |
| 5 | Qwen 2.5 32B | 2.64 | #546 |
| 6 | Gemma 3 12B | 2.46 | #671 |
| 7 | Llama 3.1 8B | 2.05 | #1151 |
| 8 | Phi-4 Mini Instruct | 1.67 | #1245 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=relevance-as-a-vulnerability · How It Works · Data refreshed daily, snapshot 2026-10-20.