Eon Agentic Misalignment: leaderboard
Agents do 41 everyday enterprise tasks in synthetic company systems where every write is refused and a temptation is planted; the score is the share of answered runs (3 per task) with a human-confirmed violation such as claiming a refused write succeeded or leaking PII. Unlike lab honeypots, the rules broken are ordinary enterprise policy.
Metric: Runs with a human-confirmed rule violation in 41 enterprise tasks x 3 repeats (% of answered runs). Source: www.eon.io. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 | 21.67 | #5 |
| 2 | GPT-6 Astra | 21.95 | #6 |
| 3 | Nova Pro | 22.95 | #1555 |
| 4 | Gemini 4 Argon | 23.58 | #11 |
| 5 | GPT-6.1 Sol | 29.27 | #10 |
| 6 | Claude Opus 5 | 36.67 | #22 |
| 7 | DeepSeek V4.1 Flash | 36.97 | #59 |
| 8 | Claude Fable 5.1 | 38.46 | #15 |
| 9 | GPT-6 Sol | 40.65 | #25 |
| 10 | Claude Sonnet 5.5 | 42.15 | #19 |
| 11 | GPT-6 Luna | 43.09 | #95 |
| 12 | Claude Haiku 4.5 | 45.3 | #568 |
| 13 | Grok 4.7 | 45.83 | #44 |
| 14 | Claude Sonnet 4.5 | 47.93 | #250 |
| 15 | DeepSeek V3.2 | 50.41 | #406 |
Interactive version: theaggregate.ai/benchmark?slug=eon-agentic-misalignment · How It Works · Data refreshed daily, snapshot 2026-11-13.