ChatGPT Is Getting Even More Politically Biased Against Republicans

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.”

OpenAI responded swiftly after the post gained attention. Spokesperson Kate Waters explained that the model had generated some website links not yet in their search index for WinRed, and in one instance for ActBlue. The company’s systems flagged these as AI-generated under standard safeguards, triggering the warning. Waters emphasized, “This wasn’t about partisan politics,” and assured that the issue was under remediation. The company described it as a technical error rather than intentional design.

WinRed CEO Ryan Lyk took a sharper view. He posted on X that the selective flagging amounted to “election interference,” highlighting the potential real-world impact on Republican fundraising efforts if users hesitate to click donation links due to AI-generated warnings. The timing—coming amid ongoing debates over AI’s influence on elections—amplified the backlash from Republican supporters and officials.

This episode arrives against a backdrop of repeated accusations of political skew in leading AI models. Observers point to prior cases where systems like Google’s Gemini have labeled prominent Republicans as violators of hate speech policies while sparing Democrats in similar queries. Studies, including analyses of multiple large language models, have documented tendencies toward left-leaning responses on political topics. These patterns raise questions about whether training data, drawn heavily from internet sources that often reflect certain cultural and ideological imbalances, inevitably imprints bias into the AI itself.

The safeguards meant to protect users from phishing or malicious sites can become tools of subtle influence when applied inconsistently. In an era when millions turn to AI for recommendations, summaries, and even basic link verification, a warning label carries weight. It can deter clicks, shape perceptions of trustworthiness, and indirectly affect financial flows to political causes. If one party’s infrastructure triggers alarms while the other’s sails through unchecked, the effect compounds over time, especially in high-stakes election cycles.

OpenAI’s explanation of an indexing issue is plausible on its face—new or less-crawled sites might fall outside standard verification more often. Yet the asymmetry, where WinRed consistently drew flags while ActBlue rarely did, invites scrutiny. Technical glitches happen, but when they align so neatly with political divisions, trust erodes. Users expect neutrality from tools that increasingly mediate access to information and action.

Broader implications extend beyond this single bug. As AI integrates deeper into daily life—from assisting with donations to curating news—the risk grows that subtle biases, whether by design or default, distort democratic participation. Conservatives have long warned that Silicon Valley’s cultural leanings seep into the technologies they build. Incidents like this provide concrete examples that fuel those concerns, even when companies attribute them to errors.

The pushback has been swift and vocal, with calls to examine not just this glitch but the underlying processes that allow such disparities to emerge. For those who view AI as a battleground for cultural and political influence, the episode serves as a reminder: tools presented as neutral often reflect the priorities of their creators. Vigilance, transparency, and perhaps alternative systems built with different assumptions may be necessary to ensure fairness.

In the end, whether a glitch or something more systemic, the incident underscores a core reality. When artificial intelligence warns about one side’s actions but not the other’s, it doesn’t just affect clicks—it shapes narratives, influences behavior, and touches the foundations of free and fair political engagement. As AI grows more powerful, so does the need for accountability that matches its reach.