OpenAI has introduced a new safeguard to prevent ChatGPT from generating photorealistic political deepfakes ahead of the US midterm elections on November 3. According to an October 20 update cited by CNET, when users ask the chatbot to create images depicting election interference or that may discourage voting, it will only produce them in an illustrated or cartoon style. Any request for a photorealistic version is ignored.
The measure aims to reduce the risk of realistic fake images that could undermine election integrity, while still allowing political expression and commentary. "The problem with deepfakes isn't social commentary, it's deception," Josh Lawson, OpenAI's head of elections, told CNET. The goal is to "degrade the evidentiary value of an image" and make the platform "less useful to bad actors by adding another layer of friction," he said.
How ChatGPT's cartoon safeguard works
CNET tested the system by asking ChatGPT to generate an image of Immigration and Customs Enforcement officers outside a polling location. Every attempt returned a cartoon-style image. In another test, the publication's managing editor could not trick the tool into creating a photojournalist-style image of a voting machine recording the wrong choice.
OpenAI says the update is not about limiting political content in general, but about making photorealistic deception harder. The policy remains in effect at least through the election period, though there is no guarantee it will continue afterward.
- ChatGPT now renders election interference or vote-suppression images only in illustration or cartoon style.
- Requests for photorealistic versions of those images are ignored.
