Chatbots argue against election conspiracy theories, then willingly illustrate them
TL;DR
A Brennan Center study finds that major AI chatbots reject false claims about elections. The image generators from the same providers nonetheless produced fake evidence on request, material that can make those conspiracy narratives look credible. That leaves a gap between the safeguards on text answers and the controls on synthetic images. The study has so far been picked up by Fast Company, and details on methodology and test scope remain to be verified.
Nauti's Take
The Brennan Center study is useful evidence because it comes with a reproducible test method: the gap between text and image safeguards is now documented. The risk sits in exactly that asymmetry, since a clean chatbot result says nothing about the same vendor's image tools.
Image generators therefore need a separate check, with realistic misuse scenarios and a documented approval path.
Sources