With the widespread use of AI-generated content (AIGC) on social media, its potential to spread misinformation poses threats to the public. Although AI disclosure is widely promoted as a transparency measure to prompt critical evaluation, its effectiveness in science communication remains controversial. This study conducted a within-subjects experiment (N = 433) to examine how AI disclosure affects perceived credibility of science communication texts and the moderating roles of readers' negative attitudes towards AI and audience involvement. The experiment manipulated AI disclosure labels and information veracity. The results revealed a truth-falsity crossover effect: AI disclosure significantly reduced the perceived credibility of correct information while unexpectedly increasing the perceived credibility of misinformation. Negative attitudes towards AI significantly moderated these effects, whereas audience involvement exerted only limited influence. These findings highlight the complex and sometimes counterproductive consequences of AI disclosure in science communication and suggest implications for cue-based processing, algorithm aversion, and the design of disclosure mechanisms.