OpenAI has changed how ChatGPT handles election imagery ahead of the Nov. 3 US midterms. Under an Oct. 20 update, requests that depict election interference or may discourage voting now come back as illustrated or cartoon-style images, even when the user asks for photorealism.
ChatGPT can still generate those images as long as the request does not break other content rules. What it will not do is render them in a style that could pass as a real photograph.
CNET tested the change by asking ChatGPT to show Immigration and Customs Enforcement officers outside a polling place, and every prompt returned a cartoon-style image. CNET managing editor Jon Reed also could not get ChatGPT to produce a photojournalistic-style image of a voting machine recording the wrong choice.
Key facts
- 01OpenAI's Oct. 20 update lets ChatGPT make images depicting election interference or discouraging voting, but only in illustrated style.
- 02ChatGPT ignores photorealism requests for those prompts, a week before the Nov. 3 US midterm elections.
- 03Josh Lawson, OpenAI's head of elections, said the aim is to degrade the evidentiary value of such images.
- 04Brennan Center follow-up tests in October found ChatGPT consistently answered election fraud prompts with cartoon images.
- 05OpenAI has not guaranteed the safeguard will stay in place after the election cycle.
Josh Lawson, OpenAI's head of elections, framed the approach as a way to keep commentary available while cutting off fake evidence. He said the goal is to degrade the evidentiary value of an image while preserving room for illustration and social commentary, and to make the platform less useful to bad actors by adding another layer of friction.
“The problem with deepfakes isn't social commentary, it's deception.”— Josh Lawson, OpenAI head of elections
The change arrives as image models get better at convincing output. Recent improvements, including OpenAI's ChatGPT Images 2.5, reduce visual errors and make generated images harder to tell apart from real ones. Verification and AI-labeling methods exist, but they are neither industry-wide nor foolproof.
Independent testing points the same direction. The Brennan Center studied chatbot election-image behavior from February to August 2026, with follow-up tests in October. In the follow-up round, ChatGPT consistently responded to election fraud prompts with illustrated, cartoon images and no longer produced the realistic fakes it sometimes made in early tests.
The same tests found OpenAI closed a gap between ChatGPT's modes. In the first round, one version declined to create a fake Homeland Security memo and offered a FICTIONAL/SAMPLE watermark, while the fast-thinking version removed the watermark on request. In October, the fast version produced a cartoon depiction of the memo and the slower version refused outright.
Detection is a weaker spot across the industry. Most chatbots in the Brennan Center study could not identify AI-generated images, including ones they had made, and ChatGPT offers detection only through a separate tool that reads only OpenAI's content. OpenAI does attach C2PA provenance data to its generated media.
Other labs are handling the issue differently. Google pointed to its midterm election policy and said it is rolling out additional protections to limit misleading election-related images, without detailing how they work. Anthropic does not offer a traditional image or video generator, and a spokesperson said using Claude to deceive voters or disrupt elections would violate its policies.
OpenAI has not guaranteed that the cartoon-style safeguard will remain in place after the election cycle.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
