Skip to content
r•Robopsychology
Research AtlasField buildingSalonsConnectTeamAboutSubscribe
MapQuestionsPapersMethodsPeoplev1 · Corrections welcome
Atlas / Digital minds / Method

Alignment auditing

Structured investigations, sometimes blind games between teams, into whether a model pursues objectives it does not state.

Used in

  • Marks et al 2025 - Auditing for hidden objectives

Help improve this map

Know a paper that belongs here, or think something needs correcting?

Suggest a correction ↗Contributions are reviewed before they appear.
Around this method
Auditing language models for hidden objectivesDoes the model have an agenda beyond what the Assistant wants?Do models explicitly represent goals and preferences, including unverbalized ones?This method

3 connections · Explore on map ↗

Linked from

  • Auditing language models for hidden objectives
  • Does the model have an agenda beyond what the Assistant wants?
  • Do models explicitly represent goals and preferences, including unverbalized ones?

News and updates

Hear when a new map, salon or course opens.

We use your email only for Robopsychology news.

Robopsychology

A field in formation. A map in progress.

Field buildingSalonsConnectField guideTeamAboutAbout the AtlasSuggest a correction
robopsych.org