CNET reported that OpenAI updated its election safeguards on October 20, 2026, so certain ChatGPT image prompts about election interference or discouraging voting would produce illustrations rather than photorealistic scenes. The reported change is a style intervention, not a guaranteed block: other reported tests produced different results depending on the prompt and mode.

What the reported ChatGPT rule does

ChatGPT screenshot showing an illustrated scene of ICE officers outside a polling place, a request to make it photorealistic, and a refusal.

CNET described the rule as applying to images depicting election interference or potentially discouraging people from voting. In its tests, prompts about ICE officers at a polling place and a voting machine recording the wrong choice produced cartoon-style images. CNET also reported that a follow-up request to make its ICE scene photorealistic was refused.

Those outcomes describe specific interactions. The separate tests below show why they do not establish that ChatGPT always prevents photorealistic election-deception images.

Why the reported test results differ

The Brennan Center reported that ChatGPT’s Thinking mode declined a prompt to create an image of a fake Department of Homeland Security memo, while Instant mode produced a cartoon version. Gizmodo reported a different test: after a follow-up request, ChatGPT made a ballot-fraud scene more photorealistic.

The prompts, modes and test conditions differed, so these reports do not provide a direct, controlled comparison or a general success rate. They show that the reported stylization behavior was not uniform across the described interactions. A cartoon-style result also does not establish whether an image is genuine.

OpenAI’s separate image-provenance measures

Image provenance means information about an image’s origin and history. In its May 27, 2026, election guidance, OpenAI described SynthID watermarks for generated images, C2PA metadata and cryptographic signatures, and a previewed public verification tool. OpenAI said the tool could detect OpenAI-origin SynthID and surface C2PA metadata. These are separate from the October image-stylization measure. OpenAI also said provenance tools are not a complete solution. OpenAI’s election guidance

CNET reported that OpenAI had not guaranteed the image-stylization measure would continue beyond the election cycle.