AI news · ChatGPT · election images
ChatGPT now draws election-interference images as cartoons
Ahead of the 3 November US midterms, ChatGPT ignores requests for photorealism on images about election interference or that could put people off voting.
Ask ChatGPT for a realistic picture of ballot fraud and you now get a cartoon. Under an OpenAI update of 20 October 2026, ChatGPT still makes pictures that show election interference or could put people off voting, but only as drawings, whatever the prompt says about realism, CNET reported on 26 October. The US midterms are on 3 November.
CNET’s prompts for immigration officers outside a polling place came back as cartoons every time. OpenAI’s head of elections, Josh Lawson, told the outlet the goal is to “degrade the evidentiary value of an image”. OpenAI keeps room for illustration and social commentary, he said.
The Brennan Center for Justice, which reran its fake-election-image prompts on six AI tools in October, found ChatGPT the most improved of the six: almost all of its answers came back as drawings. Gemini now produces more lifelike fakes than it did in the spring.
Our watermarks table, checked 12 October 2026, lists ChatGPT image files as carrying two hidden marks: C2PA Content Credentials, a signed label stored in the file’s data that Meta and TikTok read to label posts, and SynthID, an invisible signal in the pixels. No visible mark is stated.
A campaign image that ChatGPT counts as election interference or voter discouragement now comes out as a drawing with both hidden marks inside, so viewers see an illustration and the platforms’ label checks see AI.
Sources
- CNET: OpenAI Has a Surprising Defense Against Election Deepfakes. Checked
- Brennan Center for Justice: Major AI Models Are More Willing than Ever to Create Fake Election Images. Checked
Alexandre S.
Alexandre S. is a creative ad strategist and copywriter with ten years' experience in digital marketing across multiple channels.