Palisade Research - Footnote 15
The article does not mention the experiments were released in October 2025. The article states that Grok 4 went from refusing instructions from 93 percent to 97 percent of the time, not that it showed 93-97% resistance rates after stronger prompts. The article states that OpenAI's o3 model had a 23 percent shutdown resistance, not that it continued to resist shutdown even under clarified instructions designed to eliminate ambiguity.
Our claim
entire recordNo record data available.
Source evidence
1 src · 1 checkNoteThe article does not mention the experiments were released in October 2025. The article states that Grok 4 went from refusing instructions from 93 percent to 97 percent of the time, not that it showed 93-97% resistance rates after stronger prompts. The article states that OpenAI's o3 model had a 23 percent shutdown resistance, not that it continued to resist shutdown even under clarified instructions designed to eliminate ambiguity.