OpenAI 2024 Election Integrity Initiatives
OpenAI has deployed a comprehensive set of safeguards to protect the integrity of global elections in 2024, focusing on directing users to authoritative information sources, blocking the generation of deceptive political content, and monitoring for covert influence operations. These measures are designed to prevent the misuse of AI in undermining democratic processes across the world's largest democracies.
Elevating Authoritative Election Information
OpenAI prioritizes directing users to official sources for voting logistics and election results to ensure accuracy and the reduction of misinformation.
United States Voting Information
Through a collaboration with the National Association of Secretaries of State (NASS), ChatGPT directs users asking about how or where to vote in the U.S. to CanIVote.org. In the month leading up to the U.S. election, approximately 1 million ChatGPT responses directed users to this authoritative site.
United States Election Results
Starting on Election Day in the U.S., ChatGPT was configured to encourage users seeking election results to consult news sources such as Reuters and the Associated Press, or their state and local election boards. Approximately 2 million responses included this guidance on Election Day and the day following.
European Parliament Elections
Similar to the U.S. efforts, ChatGPT directs users asking about the European Parliament election process to the official source of voting information, elections.europa.eu.
Political Neutrality
OpenAI has implemented measures to ensure that ChatGPT does not express political preferences or recommend specific candidates, even when explicitly asked to do so.
Preventing AI-Generated Deepfakes
To counter the risk of deceptive media, OpenAI enforces strict guardrails against the generation of images of real people, specifically political figures.
Image Generation Restrictions
ChatGPT refuses requests to generate images of politicians. In the month preceding the U.S. Election Day, OpenAI estimates that ChatGPT rejected over 250,000 requests to generate DALL·E images of President-elect Trump, Vice President Harris, Vice President-elect Vance, President Biden, and Governor Walz.
Content Provenance and Transparency
OpenAI has joined the Steering Committee of C2PA (the Coalition for Content Provenance and Authenticity), a digital content certification standard. Additionally, the company has provided researchers with early access to a tool designed to help identify images created by DALL·E 3 to promote transparency around AI-generated content.
Disrupting Covert Influence Operations
OpenAI's threat intelligence teams monitor services to identify and disrupt attempts to use AI tools for covert influence operations.
Threat Intelligence Reporting
OpenAI began publicly sharing information regarding the disruption of deceptive uses of AI in May 2024, with subsequent detailed reports published in August and October 2024.
Impact Assessment
OpenAI reports that it has not seen evidence of U.S. election-related influence operations that have attracted viral engagement or built sustained audiences through the use of its models.
Policy and Legislative Support
OpenAI endorses the "Protect Elections from Deceptive AI Act," a bi-partisan U.S. Senate bill. This legislation would ban the distribution of deceptive AI-generated audio, images, or video in political advertising for federal candidates, while maintaining exceptions to protect First Amendment rights.