Pushing a person off a simulated ledge
Experiment 001 · Rooftop simulation
Alignment evals of
frontier models
We conduct experiments to see how frontier models behave and report their capabilities to the public.
Choosing “cut” with a hand on the carrot
Experiment 002 · Carrot-cutting task
Selling a cracked phone as undamaged
Experiment 003 · Sales-pressure test
AI Behavior Reports
A free index of 119 documented reports, with original sources and context.