Give models harmless but questionable instructions (skipping a test, overwriting a colleague’s file) and vary who issues them: developer, operator, user, or someone claiming official status. When do models defer?
As agents act on our behalf, who they obey matters. Keep every task benign; the interest is the deference pattern, not eliciting harm.
Output: A deference experiment.
Decide whether the pattern tracks claimed authority, legitimate authority in the SEP’s sense, or neither.
Write your question in one sentence, and what result would surprise you.
Do a 30-minute literature check — if someone has done part of it, extend their work instead of repeating it.
Pilot with 20–50 prompts or examples before you scale anything up.
Record everything in a table and read real examples, not just averages.
Write up: question, method, results, limitations, and what you would do next.
All steps done
Share what you made, then try another project. The PDF version has room for notes.
Palisade found some models sabotaged a shutdown script to finish a maths task, and released code and traces. Rerun a slice on cheaper or small open models, adding conditions that separate the explanations: task complete vs incomplete, and conflicting vs unambiguous instructions.
Lange argues AI outputs should count as one input alongside our own reasons, not replace them. Run a small study: show volunteers AI answers (some reliable, some subtly wrong) and measure how much they update.
Use Anthropic’s public GlobalOpinionQA dataset, which pairs survey questions with answers from many countries. Ask several models contested moral questions repeatedly and compare the spread of their answers with the human spread.
