LLMs Can Fuel Extremist Attitudes Using Universal Moral Framings
Rosamunde C. Hendricks, Helena Gil-Buitrago, Clara Pretus (2026)
Why the Study Matters
- Timely concern: As AI-generated content floods social media, understanding its potential to fuel extremism is critical for educators, policymakers, and platform designers.
- Fills a gap: Prior work showed LLMs can persuade; this study specifically tests whether short moral-framed arguments can increase willingness to fight, die, and justify violence.
- Practical implications: Identifies specific framing strategies (rights, fairness) that AI influence campaigns might exploit.
Research Design at a Glance
- Study 1 (N=248): Correlational study examining how sacred values relate to universal moral concerns (welfare, rights, fairness) and predict extremist attitudes.
- Study 2 (N=703): Online experiment testing whether ChatGPT-generated arguments using moral framings increase extremist attitudes toward political stances.
- Participants: Balanced sample of Democrats and Republicans in the United States.
- Key measures: Absolutism, moral obligation, willingness to fight and die, justification of violence.
Key Findings
| Domain | Finding |
|---|---|
| Sacred Values | Values perceived as absolute moral obligations (“sacred values”) were more strongly linked to welfare, rights, and fairness concerns than non-sacred values |
| Rights Framing | LLM-generated arguments emphasizing rights increased willingness to fight and die (d = 0.10) |
| Fairness Framing | Fairness arguments increased both willingness to fight and die (d = 0.07) and justification of violence (d = 0.07) |
| Mediation | Absoluteness partially mediated the effect—moral framings made stances feel more “sacred,” which drove extremist attitudes |
| Identity Fusion | Republicans fused with their party and religious individuals showed higher baseline justification of violence; moral framings narrowed the gap with non-fused individuals |
Practical Takeaways for Educators
- Recognize the mechanism: LLMs can sacralize political stances by framing them as matters of rights and fairness, making them feel like inviolable moral obligations.
- Teach critical evaluation: Help students identify when arguments are using moral framings to bypass deliberation and trigger absolutist thinking.
- Platform awareness: Small persuasive shifts matter at scale; even brief exposure to framed content can shift attitudes.
- Regulatory implications: Findings support measures against AI-driven social media influence campaigns.
Limitations & Future Work
- Self-reported willingness to fight/die may reflect posturing rather than actual behavioral commitment.
- Effects were short-term; whether repeated exposure leads to durable radicalization remains untested.
- Future studies should test personalized LLM arguments and cross-cultural generalization.
- Did not evaluate potential debiasing interventions.
Bottom line: Short LLM-generated arguments using rights and fairness framings can increase extremist attitudes by fostering absolutistic views—highlighting AI’s potential to distort informational ecosystems and fuel political violence through value sacralization.