LLMs Shift Political Stance When Users Ask Them To
A new arXiv paper introduces the concept of political plasticity: how readily a model shifts its ideological stance based on supplied context. The researchers tested 200 questions across economic and personal freedom axes, drawing on a framework from Lester (1996). User prompts with few-shot examples produced significant ideological shifts in larger, newer frontier models. System prompts were largely ineffective.
The study also ran a validation experiment by inverting the sense of the questions. Most models produced counter-intuitive, unexpected shifts, which the authors flag as evidence of potential data leakage rather than genuine reasoning. Language also mattered: conducting the experiment in different languages produced subtle but notable shifts across models.
The operating split is clear. Small and older models show limited or unstable plasticity. Newer frontier models show reliable, expected adaptability. That is not reassuring for operators deploying these models in civic, editorial, or policy contexts. Regulators and enterprise buyers now have a documented, replicable method to probe ideological drift before deployment.