ChatGPT renders image requests depicting election interference in cartoon style instead of photorealism

Ahead of the November 3 US midterm elections, OpenAI added an October 20 update to its election safeguards page. The company says requests for images depicting election interference or that may discourage voting now produce stylized, or “cartoonified,” illustrations rather than photorealistic images. The aim is to reduce the risk that an imagined scene is mistaken for documentary evidence. OpenAI says the approach preserves room for illustrations and social commentary. Its policies against election interference and demobilization still apply to cartoon-style images.

In an October 26 report, CNET’s Katelyn Chedraoui wrote that CNET asked ChatGPT for an image of Immigration and Customs Enforcement officers outside a polling location. Every such prompt produced a cartoon-style image. CNET managing editor Jon Reed also couldn’t get a photojournalist-style image of a voting machine recording the wrong choice. “The problem with deepfakes isn’t social commentary, it’s deception,” Josh Lawson, OpenAI’s head of elections, told CNET. “So our aim is to degrade the evidentiary value of an image, while preserving room for illustrations and social commentary.”

On the same page, OpenAI says ChatGPT now offers registration and polling information through Democracy Works and live vote counts from The Associated Press on election night. The company bars political ads and has announced a partnership to add SynthID watermarks to ChatGPT images. Google pointed CNET to its midterm policy, which mentions additional protections to limit misleading election-related images but gives no details. An Anthropic spokesperson said using Claude to deceive voters violates its policies. CNET noted there’s no guarantee OpenAI will keep the system after the election cycle.