OpenAI’s ChatGPT Flags Republican Links as Unsafe
OpenAI announced on Friday that a technical issue led ChatGPT to mark links to Republican websites as potentially unsafe for users.
This glitch particularly affected links to WinRed, which is the official donation site for the Republican Party. Interestingly, no similar warnings seemed to apply to links from ActBlue, the main fundraising platform for the Democratic Party.
Digital marketer Mike Morrison commented, highlighting the inconsistency: “ChatGPT flagged WinRed links as potentially insecure.” He noted, “Of course, ActBlue links are just fine.” This discrepancy has drawn attention, showcasing a curious flaw in the system.
Prime Minister Scott Morrison had recently encouraged ChatGPT to generate links for various political campaign merchandise stores for both the Democratic and Republican parties. However, when he clicked on the link to the GOP store hosted by WinRed, a warning appeared stating, “Please make sure this link is safe.”
Accompanying the warning were additional notes: “This link has not been verified and may lead to third-party sites. Please ensure you trust this link before proceeding.”
In contrast, Morrison found no such warning when accessing a link to a store managed by ActBlue.
OpenAI responded to the incident, stating that “this should not have happened,” and reassured that improvements are underway. Kate Waters from OpenAI elaborated, saying they immediately reached out to Morrison upon noticing the situation. She emphasized, “This is not about partisan politics.” According to her, the AI generated links for both WinRed and one instance of ActBlue that weren’t previously in the search index, resulting in the safety warnings being triggered.
Waters concluded, “This issue is now being fully resolved.”
OpenAI has faced ongoing criticism regarding perceived biases in its AI, especially leaning towards Democratic views. The warning about WinRed has sparked frustration among Republicans on social media. WinRed CEO Ryan Rick remarked, “This is election interference.” Meanwhile, Senator Mike Lee (R-Utah) questioned whether ChatGPT was inadvertently assisting Democrats.
