WHEN THE MACHINE SPEAKS SCIENCE: A CRITICAL REVIEW OF GENERATIVE-AI DISINFORMATION, ANTI-SCIENCE COMMUNITIES, AND COMMUNICATIVE RESILIENCE

Abstract

Generative artificial intelligence has made it cheap, kind of easy to produce text, images, audio and video that look real, but really are not. And this kind of capability comes in at a time when public trust in science is already kind of contested, plus organized anti-science groups have gotten pretty good at using digital platforms, with a surprising amount of skill. This review brings together three bodies of scholarship that are usually kept apart: research on AI-generated disinformation, research on anti-science movements and conspiracy communities, and research on communicative resilience. The aim is to understand what happens when synthetic content enters the contest over scientific authority, and what protects publics and institutions when it does. Drawing on a critical integrative method, the review examines literature published mainly between 2021 and 2026 across communication, science communication, and the wider social and behavioural sciences. What’s been happening in 2026 across communication, science communication and the wider social and behavioural sciences, there are a few findings that jump out. First, the danger from generative AI isn’t mainly about any one fake thing, like a single fabricated report, it’s more about the scale and speed, and the personalising part it enables, and also the messy uncertainty it spreads even when no one is fully fooled. Second, a lot of the defenses scholars rely on most—media literacy, fact-checking, psychological inoculation—were mostly designed and tested against human-made, text-based falsehood, so it’s still an open problem how well those same tools work when the media is high-quality synthetic. Third, work on citizen level resilience and work on how scientific institutions protect their reputation have been running in almost separate lanes, even though one synthetic attack can test both at once, in the same moment. The review argues that the field needs an integrated, multi-level account of communicative resilience, one that connects the psychology of individual judgement, the design of corrective messages, and the strategies of institutions, and that pays particular attention to AI-generated scientific disinformation and to lower-resilience media environments in Southern Europe. It closes by setting out a research agenda, practical implications for communicators, and policy implications tied to the European Union's emerging regulatory framework