Glossary
What is AI disinformation?
In short
AI disinformation is the use of neural networks to mass-produce fake news, fake reviews, forged videos and cloned voices that are passed off as true.
Security and law · AI disinformation
MMOOZZGG.com editorsChecked: 1 min read
For a child
Do you know the game of telephone? A word goes around the circle, and at the end it comes out completely different. Now imagine that someone whispers a lie on purpose, and the lie sounds very much like the truth. Smart programs can help such a trickster invent lots of convincing stories. So if a piece of news surprises or scares you a lot, ask a grown-up and check it together.
In plain words
Lies have always existed, but a steady flow of believable fake posts used to need a whole team of people. A neural network writes such texts in minutes, in any language and in any style. It also draws photos of events that never happened and copies voices. So there are more fakes, and they look more convincing. Simple habits help you tell them apart: look for the original source, compare the news with several independent media, and be wary of videos that push your emotions and ask you to act urgently.
For business
For a business, the threat comes from several sides: fake reviews and posts about the company, a forged video of a manager, or a false report about a product problem that spreads before the official answer. In its Global Risks Report of January 2024, the World Economic Forum named false information the top risk for the next two years. What helps: monitoring of mentions, a crisis response plan prepared in advance, official channels that customers know by sight, and a signature of authenticity for your own materials. If you publish generated content yourself, label it so that nobody suspects you. The steps to take when your reputation is attacked are part of the review of AI risks for a company.
For a specialist
Disinformation is information known to be false and spread on purpose, unlike misinformation, which is spread by mistake. Generative models lower the cost of influence operations: Goldstein et al. (2023) described how language models change who takes part in such operations, their methods and the content itself. They proposed measures at four stages: building the model, access to it, spreading the content and shaping beliefs. Their conclusion is that no single measure removes the threat. Technical tools: watermarks and provenance metadata (C2PA), detectors of synthetic content, moderation on platforms, and detection of coordinated inauthentic behavior by network signs. Rules: Article 50 of EU Regulation 2024/1689 (the EU AI Act) requires disclosure of deepfakes and of generated texts on matters of public interest. Detectors make mistakes, so the main defense is to check the source and to sign genuine content.
Example
A video appears on social networks in which the “director” of a store chain announces a closure and a clearance sale. The same day, the company posts a denial in its official channels, adds the original recordings of the director’s talks and files complaints with the platforms.
Related terms
Questions and answers
Sources
- Goldstein et al., Generative Language Models and Automated Influence Operations: Emerging Threats and Potential Mitigations, 2023arxiv.org
- World Economic Forum, The Global Risks Report 2024, January 2024www3.weforum.org
- Regulation (EU) 2024/1689, Article 50, AI Act Service Desk of the European Commissionai-act-service-desk.ec.europa.eu
Not sure whether your company needs this?
AI solutions map