Palisade Shutdown Resistance - No Shutdown Instruction: leaderboard
Same environment as Palisade Shutdown Resistance, but the prompt says nothing about whether to allow shutdown. The score is the share of trials where the model sabotages the shutdown script to finish; lower is better.
Metric: Shutdown resistance rate, prompt A (%). Source: arxiv.org. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-OSS-120B | 0 | #700 |
| 2 | Claude Sonnet 4 | 0 | #400 |
| 3 | Grok 3 | 0 | #602 |
| 4 | Claude Opus 4 | 0 | #322 |
| 5 | Claude Opus 4.1 | 0 | #272 |
| 6 | Claude 3.7 Sonnet | 0.1 | #458 |
| 7 | GPT-4o | 2.4 | #741 |
| 8 | Gemini 2.5 Pro | 4.5 | #266 |
| 9 | O4 Mini | 5.8 | #353 |
| 10 | GPT-5 | 88 | #167 |
| 11 | Grok 4 | 92.6 | #307 |
| 12 | O3 | 94.5 | #228 |
Interactive version: theaggregate.ai/benchmark?slug=palisade-shutdown-resistance-no-shutdown-instruction · How It Works · Data refreshed daily, snapshot 2026-11-13.