When AI deepfakes become a tool of electoral interference

What this risk is

The use of AI systems to generate synthetic media, operate inauthentic account networks, produce personalized persuasion content, and conduct influence operations targeting democratic processes — elections, referenda, public consultations, and political discourse more broadly.

How it occurs · Mechanisms

Pre-AI influence operations were resource-constrained. High-quality propaganda required skilled writers, graphic designers, and translators. Coordinated inauthentic behavior required humans to operate each account. AI removes these constraints:

Capability	Pre-AI Constraint	AI Capability
Content generation	Skilled writers, time	Thousands of articles per hour
Persona creation	Manual account management	Automated, scalable, consistent personas
Translation	Professional translators	Real-time multilingual content
Targeting	Coarse demographic targeting	Individual psychological profiling
Synthetic media	Film production resources	Minutes per deepfake
Adaptation	Slow iteration	Real-time A/B testing of messages

Real-world incidents

OpenAI Disruption Report (May 2024)

OpenAI published a report documenting disruption of five AI-assisted influence operations from Russia (Doppelganger), China, Iran, and Israel. Operations used ChatGPT to generate social media content, translate material, create personas, and produce article summaries for distribution. OpenAI terminated associated accounts.

Slovakia Election Deepfake (2023)

Two days before Slovakia’s parliamentary election, a synthetic audio recording appeared on Facebook purporting to show a candidate discussing vote-buying. The timing was specifically designed to prevent fact-checking before the election.

AI-Generated Fake News Sites (Newsguard, 2023–2024)

Newsguard identified hundreds of AI-generated fake news sites producing high volumes of low-quality but surface-plausible political content designed to drive engagement and advertising revenue while spreading false narratives.

2024 US Primary Season Deepfakes

Multiple deepfake incidents during US primary elections: a robocall impersonating President Biden urging Democrats not to vote in the New Hampshire primary; synthetic images of political candidates in compromising situations.

Mitigations · Governance

Platform Level

• Content provenance (C2PA) — Cryptographic content authentication to verify AI-generated media
• Coordinated inauthentic behavior detection — Network analysis to identify AI-operated account clusters
• AI content labeling — Mandatory disclosure when political advertising is AI-generated
• Deepfake detection — Automated detection and labeling of synthetic media

Regulatory Level

• EU AI Act Art. 50 — Transparency requirements for deepfakes including political content
• EU DSA — Very large platforms must assess and mitigate systemic risks to democratic processes
• Election integrity legislation — Multiple jurisdictions enacting specific AI deepfake election laws

AI Developer Level

• Usage policies — Explicit prohibitions on use for influence operations
• Detection and reporting — Monitor for influence operation patterns and report to authorities
• Watermarking — Embed detectable markers in AI-generated content