OpenAI publishes its approach to 2024 worldwide elections

OpenAI published a set of policies and tools aimed at limiting misuse of its products during a year with major elections scheduled in dozens of countries, including the US presidential race. The company said it did not allow applications built for political campaigning or lobbying, barred chatbots that impersonated real candidates, and prohibited uses that could discourage voting or misrepresent how or where to vote.

For US users, OpenAI said it had partnered with the National Association of Secretaries of State so that ChatGPT would direct voting-related questions to CanIVote.org, a nonpartisan clearing house for official state election information, rather than attempt to answer them directly — an acknowledgement that the model’s own knowledge of voting logistics, which vary by state and change over time, could not be trusted to stay current or accurate.

On the image side, OpenAI said DALL-E 3 outputs would carry digital credentials, developed with a cross-industry coalition, encoding metadata about their AI origin, and that it was testing a “provenance classifier” able to detect whether an image had been generated by its models, including after common edits. The company planned to give journalists, platforms and researchers early access to the classifier for feedback rather than releasing it publicly at once.

The announcement arrived alongside similar moves from other developers and platforms that year, including Anthropic’s own election safeguards the following month. It set expectations for what a “responsible” election posture from a major AI lab looked like — largely disclosure, redirection to authoritative sources and use-policy restrictions — well before votes were cast, though it offered no guarantee against synthetic content produced by other, less constrained tools.