OpenAI blocks photorealistic political deepfakes in ChatGPT

Ahead of the US midterms on Nov. 3, OpenAI now renders requests for images depicting election interference or discouraging voting only as illustrations or cartoons. Photorealistic versions of such prompts are ignored to strip images of evidentiary value for bad actors.

• ChatGPT renders election-related prompts only in cartoon style
• Photorealistic versions of such images are refused
• The safeguard targets the US midterms on Nov. 3
• Google and Anthropic did not detail their own measures