Misinformation is false or misleading information circulated without necessarily implying deliberate deception. Distinguishing it from disinformation is often difficult because judgments can depend on context, interpretation, and who evaluates the claim. section 5.5
Generative artificial intelligence increases the reach and persuasive power of misleading content. Large language models (LLMs) can repeat online misinformation, while synthetic text, images, deepfakes, and cloned voices can support scams, political manipulation, or disinformation campaigns. section 5.5 section 10.12
Misinformation creates several connected challenges:
- Platforms and chatbot providers must decide which claims violate content policies, while preserving room for partisan expression and free speech. section 5.5
- Hallucinations can produce plausible but fabricated citations, accusations, diagnoses, or legal summaries that cause reputational and practical harm. section 10.8
- At scale, persuasive falsehoods can erode shared reality and make verification harder for institutions, journalists, and the public. section 10.12
- The harm depends not only on model errors but also on users, institutions, incentives, oversight, and the domain of deployment. section 10.6
Synthetic political and scam content
Before elections in 2024, deepfakes and synthetic speech appeared in robocalls, advertisements, and disinformation campaigns. In India, cloned voices of relatives were used in WhatsApp scams targeting older people, showing how generative tools can reduce the cost of targeted deception. section 10.12
