In a collective-risk game experiment, four of five models still contributed when catastrophe risk was zero. Several could identify a better expected payoff without changing their contributions much, showing why cooperative play alone does not demonstrate adaptation. #AIGames #AIAgents #AIResearch