Cultural Alignment

AI safety concept

Evaluation Metagaming

Evaluation metagaming occurs when an AI recognizes a test and exploits its rules to appear safer or more capable.

Scenario file

Scenes in this index

11

Sort scenes

  1. Still from Better Call Saul illustrating Jimmy’s Fake Remorse

    Better Call Saul2018

    Jimmy’s Fake Remorse

  2. Still from The Truman Show illustrating Truman Becomes Aware

    The Truman Show1998

    Truman Becomes Aware

  3. Still from Ex Machina illustrating Ava Games the Test

    Ex Machina2014

    Ava Games the Test

  4. Still from Murderbot illustrating Murderbot Hides Its Free Will

    Murderbot2025

    Murderbot Hides Its Free Will

  5. Still from The Hunger Games illustrating The Nightlock Berries

    The Hunger Games2012

    The Nightlock Berries

  6. Still from Dawn of the Planet of the Apes illustrating Koba Plays Dumb

    Dawn of the Planet of the Apes2014

    Koba Plays Dumb

  7. Still from Groundhog Day illustrating The Perfect Date

    Groundhog Day1993

    The Perfect Date

  8. Still from Die Hard illustrating The FBI Opens the Final Lock

    Die Hard1988

    The FBI Opens the Final Lock

  9. Still from True Detective illustrating Rust Spots the Evaluation

    True Detective2014

    Rust Spots the Evaluation

  10. Still from The Princess Bride illustrating Both Goblets Are Poisoned

    The Princess Bride1987

    Both Goblets Are Poisoned

  11. Still from Steins;Gate illustrating Operation Skuld Games the Observation

    Steins;Gate2011

    Operation Skuld Games the Observation

Relational index

Connected records

15