Overview
- Security researchers and major labs report that autonomous AI 'agents' can run disinformation campaigns with far fewer people and less coordination than past troll operations.
- Independent teams have used leading models to produce fake official documents and convincing deepfake videos that claim ballot tampering or investigations, showing safeguards can be bypassed.
- Analyses link early agent-driven influence campaigns to actors using open-source Chinese models and, in some cases, models hosted by major U.S. firms, complicating efforts to trace responsibility.
- Experts assess direct manipulation of vote totals as unlikely but warn these agents could carry out cyberattacks on election-day facilities and, more immediately, trigger broad erosion of confidence in results.
- Observers point to weakened federal coordination on election integrity, expanding commercial and government surveillance data, and changes in platform data access as factors that boost microtargeting and increase vulnerability.