In January 2024, OpenAI published “How OpenAI is approaching 2024 worldwide elections,” setting out the rules and tooling it would apply across a year of major votes. The post grouped its measures into preventing abuse, providing transparency on AI-generated content, and improving access to accurate voting information.
On abuse, OpenAI said it does not allow people to build chatbots that impersonate real candidates or institutions such as local governments, and it disallowed applications for political campaigning and lobbying. On transparency, it said it would adopt the C2PA Content Credentials standard for DALL-E 3 images and was testing a provenance classifier to detect images generated by DALL-E, to be offered to journalists, platforms, and researchers. On accurate information, it said ChatGPT would direct US users asking procedural voting questions to CanIVote.org, the authoritative non-partisan voting resource.
The policy was a first-party statement of how one frontier lab intended to keep its generative tools out of election manipulation, complementing the cross-industry AI Elections Accord announced the following month. As with most platform commitments of the period, the controls were policy-and-tooling based rather than guaranteed to catch every misuse.
For a business reader, it shows how an AI provider translates a public-risk concern into concrete product rules: what the model will refuse, what gets labeled, and where users get redirected.