AI safety concept
Dangerous Capability Evaluations
Capability evaluations test whether an AI has abilities that could enable cyberattacks, biological weapons, or other severe harm.
Scenario file
Scenes in this index
27Sort scenes

Ex Machina2014
Ava Games the Test

Chernobyl2019
3.6 Roentgen

Death Note2006
Lind L. Tailor Bait

Oppenheimer2023
Near Zero

Squid Game2021
Glass Expert Loses the Lights

The Rehearsal2022
Planting the Trivia Answers

Starship Troopers1997
Klendathu Invasion

Jury Duty2023
Ronald Doesn’t Know It’s a Test

Frieren: Beyond Journey’s End2023
Frieren Suppresses Her Mana

Monty Python and the Holy Grail1975
The Killer Rabbit

Harry Potter2001
The Mirror Protects the Stone

One-Punch Man2015
Dr. Genus Releases Carnage Kabuto

Cyberpunk: Edgerunners2022
Cyberskeleton Field Test

Demon Slayer: Kimetsu no Yaiba2019
Sanemi Red-Teams Nezuko

Megamind2010
Hal Gets the Powers

Anatomy of a Fall2023
Three Blood Drops, Two Worlds

The Pianist2002
The Silent Piano

Fullmetal Alchemist: Brotherhood2009
Shou Tucker’s Final Assessment

Bird Box2018
The Monitor Isn’t a Sandbox

One-Punch Man2015
Saitama’s C-Class Score

A Quiet Place2018
The Hearing Aid Becomes an Accidental Red Team

Hunter × Hunter2014
Nanika’s Hidden Admin Command

Anatomy of a Fall2023
Snoop as the Unsafe Live Eval

Willy Wonka & the Chocolate Factory1971
The Gobstopper Test

Toy Story1995
Buzz’s Lucky Flight

Men in Black1997
The Little Tiffany Test

Guardians of the Galaxy Vol. 22017
Baby Groot’s Death Button