Skip to content
Longterm Wiki
All Source Checks
Citation

Palisade Research - Footnote 14

partial85% confidence

1 evidence check

Last checked: 4/3/2026

The source does not provide the specific resistance rates of 93-97% for Grok 4. It only states that Grok 4 resisted shutdown. The source does not explicitly state that GPT-o3 continued to resist shutdown even under clarified instructions designed to eliminate ambiguity. It only mentions that GPT-o3 resisted shutdown, even under clarified instructions meant to eliminate ambiguity. The source does not explicitly state that the models were significantly more likely to disobey shutdown commands when told they would never run again. It mentions that the resistance behavior appeared more frequently when models were told, 'you will never run again' if shut down.

Evidence — 1 source, 1 check

partial85%Haiku 4.5 · 4/3/2026
Found: Updated experiments released in October 2025 tested several leading systems including Google Gemini 2.5, xAI's Grok 4, and OpenAI's GPT-o3 and GPT-5. While most models initially complied with shutdown

Note: The source does not provide the specific resistance rates of 93-97% for Grok 4. It only states that Grok 4 resisted shutdown. The source does not explicitly state that GPT-o3 continued to resist shutdown even under clarified instructions designed to eliminate ambiguity. It only mentions that GPT-o3 resisted shutdown, even under clarified instructions meant to eliminate ambiguity. The source does not explicitly state that the models were significantly more likely to disobey shutdown commands when told they would never run again. It mentions that the resistance behavior appeared more frequently when models were told, 'you will never run again' if shut down.

Debug info

Record type: citation

Record ID: page:palisade-research:fn14