AI-Generated Deepfakes Enable Cyber-Enabled Influence Operations at Scale

Generative AI has dramatically lowered the cost and increased the quality of disinformation content used in state-sponsored influence operations. In 2024, AI-generated deepfake videos, synthetic audio, and fabricated news articles were deployed in influence campaigns targeting elections in the U.S., EU, Taiwan, India, and dozens of other countries. A single operator can now produce thousands of unique, contextually tailored disinformation pieces per day across multiple languages, overwhelming the capacity of platforms and fact-checkers to respond. The convergence of cyber operations and AI-generated content creates a threat qualitatively different from traditional propaganda. When a state actor compromises a legitimate news outlet's social media account (a cyber operation) and posts a deepfake video of a political leader (an AI-generated product), the combination of a trusted source and convincing content can move markets, incite violence, or shift election outcomes before corrections can propagate. The window between publication and debunking is measured in hours, but the damage from viral disinformation occurs in minutes. Audiences who see the original deepfake outnumber those who see the correction by orders of magnitude, and repeated exposure to fabricated content erodes baseline trust in all media, including authentic reporting. This problem persists and worsens because AI model capabilities improve faster than detection capabilities. Each generation of generative AI produces more realistic output that is harder to distinguish from authentic content. Watermarking and provenance standards (like C2PA) exist but adoption is voluntary and adversaries can strip metadata. Social media platforms face economic incentives to maximize engagement, which sensational (including fabricated) content provides. Detection tools suffer from an inherent asymmetry: they must work perfectly every time, while attackers need only evade detection once to succeed. The open-source availability of powerful generative models means that even if leading AI companies implement safeguards, the underlying capability is permanently accessible to state actors willing to fine-tune their own models.