AI offers the great potential to improve human lives, from problem-solving and efficiency gains to increased productivity. To date, I have not seen any sign that the political polarization that characterizes our society has affected AI. However, I believe it’s coming.

On March 5, 2025, the Arizona Republic published an opinion piece by EJ Montini, sharing the results of a question he posed to Elon Musk’s AI chatbot, Grok. He asked Grok whether the person who addressed Congress on Tuesday night, Trump’s first State of the Union, is a “Putin-compromised” Russian asset. After analyzing the transcript and public records dating back to the 1980s, the AI chatbot determined there was a 75–85% likelihood that the person who delivered the State of the Union address is a Russian asset.

The challenge here is that we all know what happens when partisans are presented with facts that are inconvenient to their ideology and beliefs. Supporters of the President will, of course, deny that he is compromised and instead argue that the problem lies with the AI—despite Musk’s claim that Grok is a “maximally truth-seeking AI, even if that truth is sometimes at odds with what is politically correct.” This type of response from AI is inevitably going to place the technology in the right’s political crosshairs.

I wonder what can be done to proactively prevent the politicization of AI. Like any other branch of science, AI is only useful if its outputs are accepted and embraced rather than rejected. We have repeatedly seen experts attacked because the facts they present clash with particular political worldviews. As a tool trained on data, we face the risk that LLMs will be retrained on propaganda or tuned to generate “alternative facts.” The discussion about this technology thus far has been focused on asking how to keep humans safe from AI. Another and just as important question is, how do we keep AI safe from humans?