Carnegie Endowment for International Peace: We Need Ongoing Monitoring of AI and Political Information

By Danaé Metaxa and Alex Engler

Generative artificial intelligence (AI), and especially large language models (LLMs), is playing an increasingly substantial role in political discourse—for example, discussing political topics, answering legal questions, and relating election information to users. LLMs produce content that regularly reaches billions of people through chatbots and search engines, and a growing number of people are using them for political information.

As we demonstrate in this article, a growing body of scientific evidence shows that LLMs are not just ubiquitous, but persuasive and influential when discussing political issues. Based on their reach and influence, we see their impact on political discourse as a critical new frontier of modern politics. Today, the world is on the precipice of a new information ecosystem, roughly analogous to the start of widespread adoption of social media two decades ago. Ensuring that the public and policymakers understand the implications of this transition and are informed to make key decisions will require extensive independent research, just as social media has.

[. . .]

Judgments pertaining to LLM refusal pose a similar challenge; responsible developers must balance potentially competing interests including freedom of expression, factual accuracy, nondiscriminatory speech, and child safety. A report from the Future of Free Speech evaluated models on their support for free expression, finding that xAI’s Grok 4 was the strongest LLM tested in terms of free-speech culture. But that same model, launched immediately following Grok 3’s infamous MechaHitler incident, is also accused of generating tens of thousands of naked images of children. Balancing freedom of expression and access to information on one hand, and safety guarantees and harm mitigation on the other, is an ongoing challenge.

Read More