AI-Generated Personas Threaten to Manipulate Election Discourse

Generative AI and large language models can now create virtual personas with consistent identities, voices and viewpoints that engage continuously in social media discussions and refine their arguments in response to feedback. Researchers say such automated influence operations, if deployed in elections, could shape public opinion cheaply and at scale. They may also be harder to detect than conventional fake accounts, threatening trust in information and democratic decision-making.

The latest research warns that AI-generated personas can operate large numbers of accounts simultaneously, test different narratives and adjust messages in real time based on users’ reactions, allowing political propaganda to spread largely unnoticed. Existing reports do not identify the research institution, publication date, test scale or any amounts involved, but researchers have classified this type of LLM-driven opinion manipulation as a core risk in the next election.