Skip to content
Longterm Wiki
Index
Citation·page:palisade-research:fn15

Palisade Research - Footnote 15

Verdictpartial85%
1 check · 4/3/2026

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 record

No record data available.

Source evidence

1 src · 1 check
partial85%Haiku 4.5 · 4/3/2026

NoteThe 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.

Case № page:palisade-research:fn15Filed 4/3/2026Confidence 85%