Reproduce “When Robots Say No: The Empathic Ethical Disobedience Benchmark” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Dmytro Kuzmenko, Nadiya Shvai
EED Gym is a simulation benchmark for testing whether agents can refuse risky instructions while preserving cooperation. It compares reinforcement-learning policies and models trust and affect alongside safety. Under stress-test evaluation, Masked PPO has 8.3% unsafe compliance versus 9.9% for vanilla PPO, but refuses more often. These results apply within the benchmark and do not establish real-world robustness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.