The retraction of scholarly articles due to errors or misconduct reflects fundamental problems in the research process and raises concerns about research integrity. However, retracted articles often continue to circulate online, especially on platforms like YouTube, which may lack effective mechanisms for academic oversight. Despite this concern, little is known about how retracted articles are actually referenced in such environments, and no comprehensive empirical investigations have been conducted to date. To investigate this issue empirically, we analyzed the prevalence and context of retracted article references on YouTube using a large-scale dataset of scholarly references extracted from video description text. We identified 1,002 references to retracted articles, corresponding to 360 unique DOIs, found in 956 videos uploaded by 714 channels. Notably, the channels referencing retracted articles accounted for nearly 1% of all channels referencing scholarly articles. Moreover, approximately 75% of the referenced articles were retracted for serious reasons such as data fabrication or ethical violations. Almost half of the videos referencing retracted articles were published after the retraction date. A substantial portion of these videos do not mention the retraction status, and several appear to reference the articles without recognizing their retracted nature, as suggested by the absence of any mention of retraction in the video content or description text. In contrast, a few videos explicitly addressed the retraction and were intended to educate viewers by highlighting issues of scientific misconduct. These findings highlight the continued circulation of unreliable scholarly information on YouTube and underscore the need for improved visibility of retraction notices in user-generated content.