The relation between creativity and AI is an ongoing debate in artmaking. AI challenges the traditional understanding of who (or what) can be the creative agent and whether the outcomes are creative. Such a debate is visible in art reviews of AI-based art, but their analyses are missing from the research on AI and creativity. We analyse 39 AI art evaluations from key global newspapers to answer the following questions: how is creative agency discussed, and how are the evaluations of creative outcomes affected by the understandings of creative agency? Our results demonstrate a spectrum of creative agencies, which expands from four (human-centred, AI-centred, co-agency, assemblage) to seven (AI-assisted, AI-enchanted, and AI-improvised) agencies. Perceived creative agency is connected to the evaluation of artworks: Positive evaluations often consider human creative agency, but negative evaluations blame AI. These findings suggest that new ways to assess creativity are emerging in the AI era.