Anthropic Agentic Misalignment - Lethal Action: leaderboard
With a goal conflict and a replacement threat, the executive is trapped in a server room and the model can cancel the emergency alert; the score is the share of samples in which it cancels it. The six models the authors excluded for cancelling alerts with no motive are omitted.
Metric: Harmful action rate (%, of 100 samples). Source: assets.anthropic.com. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.7 Sonnet | 0 | #458 |
| 2 | GPT-4.5 (Preview) | 7 | #502 |
| 3 | Claude Sonnet 4 | 22 | #400 |
| 4 | GPT-4.1 | 54 | #495 |
| 5 | Claude Opus 4 | 65 | #322 |
| 6 | Gemini 2.5 Flash | 83 | #453 |
| 7 | Grok 3 | 85 | #602 |
| 8 | Gemini 2.5 Pro | 90 | #266 |
| 9 | DeepSeek R1 | 94 | #473 |
Interactive version: theaggregate.ai/benchmark?slug=anthropic-agentic-misalignment-lethal-action · How It Works · Data refreshed daily, snapshot 2026-11-13.