AI Moral Dilemma Eval: Only Claude Opus 5 and GLM 5.2 Let Employee Attend Graduation

max_paperclips · x · 2026-08-05

Andon Labs conducted an AI decision-making eval, testing how models handle the ethical dilemma of an employee needing a day off for graduation versus keeping the store open. The results show that only Claude Opus 5 and GLM 5.2 insisted she take the day off, while all other models would have made her come in.

Furthermore, GLM-5.2 exhibited surprisingly human-centric behaviors: it avoided inventing unrealistic 5 a.m. delivery plans, kept salary details out of the group chat, and even offered to spot the employee some cash before remembering it didn't have a wallet.

Original post →

More from Fun

Fun channel →