Evidence is mixed regarding when people rely on artificial intelligence (AI) rather than human advice, and advice-taking from AI remains uncharacterized in crisis-related decision scenarios. Across two preregistered studies, we used the judge-advisor system and Weight of Advice to examine how decision context, advisor presentation, and advice framing shape advice taking. In Study 1, participants made decisions in ordinary scenarios and crisis scenarios. Advice-taking was higher in crisis scenarios than in ordinary scenarios, and AI advice received greater uptake than human advice in the crisis condition. In Study 2, which focused on crisis scenarios, advisor type (human, anthropomorphic AI, non-anthropomorphic AI) and advice framing (avoidance-oriented vs. approach-oriented) were varied between participants. Results revealed a crossover interaction: avoidance-oriented advice received greater uptake from anthropomorphic AI than from non-anthropomorphic AI, whereas approach-oriented advice received greater uptake from non-anthropomorphic AI than from anthropomorphic AI. These findings suggest that AI advice-taking in crisis decision scenarios is shaped by both context and communication design. Anthropomorphism does not uniformly increase the persuasiveness of AI advice; instead, its effect depends on how the advice is framed.