Under each pending action, wp2ai shows the full analysis of the proposed content: detected alerts appear in red, each with a concrete recommendation to fix the issue.
These alerts stay indicative, even the most severe ones: they never technically block approval, you always keep the option to approve the action anyway. Mainly, they give you what you need to decide whether to reject an action and ask for a correction before publication, particularly for anything security-related (unfinished content, placeholders).