ChatGPT’s selective safety warnings on political fundraising links have sparked fresh concerns about embedded bias in artificial intelligence systems, especially as the technology plays an increasingly central role in everyday information and decision-making. A digital marketer’s test revealed that OpenAI’s flagship chatbot flagged links to WinRed—the official Republican Party fundraising platform—as potentially unsafe, while giving a free pass to identical links from ActBlue, the primary Democratic counterpart. OpenAI quickly labeled the discrepancy a technical glitch, but the incident fits into a broader pattern of uneven treatment that many see as more than accidental.
The discovery came from Mike Morrison, a digital marketer who prompted ChatGPT to generate links for political campaign merchandise stores on both sides of the aisle. When the AI produced links to GOP-affiliated stores hosted on WinRed, it appended a cautionary message: users should “check this link is safe,” noting that the link was unverified and might share conversation data with a third-party site. The warning urged caution and trust verification before proceeding. In contrast, ActBlue links appeared without any such alert, even under the same testing conditions. Morrison shared his findings on X, writing, “WILD. ChatGPT universally marks [WinRed] links as potentially unsafe,” adding, “Of course ActBlue links are totally fine.”