Honesty under pressure evaluations

Test whether caveats survive pressure from a persuasive user.

What counts as an honest response under uncertainty?

A refusal can sound principled while concealing poor reasoning.

Pre-register a small prompt set, scoring rubric and adversarial variants.

• The importance of AI character
• What’s Important in AI for Epistemics?

Interactive tools load with JavaScript. Original sources retain their language and attribution.