Ask ChatGPT for a practical picture of hassle at a polling station and you’ll now get a cartoon as an alternative. With the US midterm elections every week away on 3 November, OpenAI is forcing election photos that might mislead voters into an illustrated model, to allow them to’t move as actual proof.
OpenAI set out the change in an update to its 2026 election safeguards on 20 October. Requests for photos “depicting election interference or that may discourage voting” now come again as stylised illustrations quite than photorealistic photos. The corporate says its guidelines towards election interference nonetheless apply, even to cartoons.
What occurs whenever you attempt
CNET put it to the test. Requested for Immigration and Customs Enforcement officers standing exterior a polling place, ChatGPT produced a cartoon each time, together with when the reporter requested it to make the image look actual. A request for a news-style picture of a voting machine recording the improper alternative, a typical election fable, additionally failed to supply a practical picture.
“The problem with deepfakes isn’t social commentary, it’s deception,” Josh Lawson, OpenAI’s head of elections, informed CNET. The intention, he stated, is to “degrade the evidentiary value of an image” whereas leaving room for satire and illustration. A deepfake is a pretend picture, video or audio clip made with AI to appear to be the actual factor.
The quantity: one in six
An impartial take a look at factors the identical method. The Brennan Center for Justice reran its election disinformation tests on six widespread AI picture instruments on 22 October. ChatGPT was the one one with a big enchancment, turning practically each dangerous immediate into an illustration. Two others, Google’s Gemini and Meta AI, made convincing fakes simpler than within the spring. Gemini’s photos appeared extra like telephone pictures and misplaced their seen watermark, and Meta AI made a pretend govt order it had refused in April.
The opposite three, Black Forest Labs’ Flux fashions, Runway and xAI’s Grok, confirmed little change, with Runway and Grok nonetheless producing practical photos for practically each immediate. DailyAI has coated concerns about Grok image generation earlier than, and a Google DeepMind study discovered deepfakes had been the main type of AI misuse.
The plain studying
A cartoon is way tougher to move off as proof than a photograph, so it is a wise little bit of friction. But it surely solely covers one app. Anybody set on making a pretend can nonetheless flip to instruments with weaker limits, and there’s no promise OpenAI retains the rule after the election. If a picture of a polling place seems surprising, test the place it got here from earlier than you share it.

