How OpenAI is approaching 2024 worldwide elections

Update on November 8, 2024:

• In the lead-up to the US elections, we implemented safeguards to direct people to reliable sources of information, prevent deepfakes, and counter efforts by malicious actors. Here are some early insights into those measures:

Elevating authoritative sources of information

• Throughout 2024, we’ve worked to elevate reliable sources of election information within ChatGPT. Through our collaboration with the National Association of Secretaries of State (NASS), we directed people asking ChatGPT specific questions about voting in the U.S., like where or how to vote, to CanIVote.org⁠(opens in a new window). In the month leading up to the election, roughly 1 million ChatGPT responses directed people to CanIVote.org⁠(opens in a new window). Similarly, starting on Election Day in the U.S., people who asked ChatGPT for election results received responses encouraging them to check news sources like the Associated Press and Reuters. Around 2 million ChatGPT responses included this message on Election Day and the day following.  

• In addition to our efforts to direct people to reliable sources of information, we also worked to ensure ChatGPT did not express political preferences or recommend candidates even when asked explicitly.

Preventing deepfakes

• We’ve applied safety measures to ChatGPT to refuse requests to generate images of real people, including politicians. These guardrails are especially important in an elections context and are a key part of our broader efforts to prevent our tools being used for deceptive or harmful purposes. In the month leading up to Election Day, we estimate 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.

Disrupting threat actors 

• A central part of our global elections work in 2024 has been identifying and disrupting attempts to use our tools to generate content used in covert influence operations. In May⁠, we began publicly sharing information on our disruptions, and published additional reports in August⁠ and October⁠(opens in a new window). 

• Our teams continued to monitor our services closely in the lead-up to Election Day and have not seen evidence of U.S. election-related influence operations attracting viral engagement or building sustained audiences through the use of our models.

Update on October 31, 2024:

• As we approach Election Day in the U.S., our teams are actively testing the safeguards we’ve put in place over the past year and monitoring for any issues or attempts to evade them. We will adjust our protective measures as needed, guided by ongoing insights into how people engage with our tools.

• Starting on November 5th., people who ask ChatGPT about election results will see a message encouraging them to check news sources like the Associated Press⁠(opens in a new window) and Reuters⁠(opens in a new window), or their state or local election board for the most complete and up-to-date information.

• This effort builds on our collaboration with the National Association of Secretaries of State to direct people looking for information about how and where to vote to CanIVote.org⁠(opens in a new window), the authoritative website on U.S. voting information. ChatGPT will continue to direct people asking these questions to CanIVote.org through Election Day.
  

Update on May 14, 2024:

• As part of our ongoing work to promote transparency around AI content during this important election year, we recently began providing researchers with early access to a new tool that can help identify images created by OpenAI's DALL·E 3. We also joined the Steering Committee of C2PA—the Coalition for Content Provenance and Authenticity. C2PA is a widely used standard for digital content certification, developed and adopted by a wide range of actors including software companies, camera manufacturers, and online platforms. 

• Building on our efforts to direct people to authoritative sources of information about voting in the U.S., we’ve introduced a new experience ahead of the 2024 election for the European Parliament. ChatGPT now directs users to the European Parliament’s official source of voting information, elections.europa.eu⁠(opens in a new window), when asked certain questions about the election process such as where to vote. This is similar to our collaboration with the National Association of Secretaries of State (NASS) for the 2024 US Presidential election.

• In addition to the steps we’re taking at OpenAI, we believe there is an important role for governments. Today we are endorsing the “Protect Elections from Deceptive AI Act⁠(opens in a new window),” a bi-partisan bill proposed by Senators Klobuchar, Hawley, Coons, Collins, Ricketts, and Bennet in the United States Senate. The bill would ban the distribution of deceptive AI-generated audio, images, or video relating to federal candidates in political advertising, while including important exceptions to protect First Amendment rights. We do not want our technology—or any AI technology—to be used to deceive voters and we believe this legislation represents an important step to addressing this challenge in the context of political advertising.

Protecting the integrity of elections requires collaboration from every corner of the democratic process, and we want to make sure our technology is not used in a way that could undermine this process. 

Our tools empower people to improve their daily lives and solve complex problems—from using AI to enhance state services⁠(opens in a new window) to simplifying medical forms for patients⁠(opens in a new window).

We want to make sure that our AI systems are built, deployed, and used safely⁠. Like any new technology, these tools come with benefits and challenges. They are also unprecedented, and we will keep evolving our approach as we learn more about how our tools are used.

As we prepare for elections in 2024 across the world’s largest democracies, our approach is to continue our platform safety work by elevating accurate voting information, enforcing measured policies, and improving transparency. We have a cross-functional effort dedicated to election work, bringing together expertise from our safety systems, threat intelligence, legal, engineering, and policy teams to quickly investigate and address potential abuse. 

The following are key initiatives our teams are investing in to prepare for elections this year:

We expect and aim for people to use our tools safely and responsibly, and elections are no different. We work to anticipate and prevent relevant abuse—such as misleading “deepfakes”, scaled influence operations, or chatbots impersonating candidates. Prior to releasing new systems, we red team them, engage users and external partners for feedback, and build safety mitigations to reduce the potential for harm. For years, we’ve been iterating on tools to improve factual accuracy, reduce bias, and decline certain requests. These tools provide a strong foundation for our work around election integrity. For instance, DALL·E has guardrails to decline requests that ask for image generation of real people, including candidates.

We regularly refine our Usage Policies⁠ for ChatGPT and the API as we learn more about how people use or attempt to abuse our technology. A few to highlight for elections: 

• We’re still working to understand how effective our tools might be for personalized persuasion. Until we know more, we don’t allow people to build applications for political campaigning and lobbying. 

• People want to know and trust that they are interacting with a real person, business, or government. For that reason, we don’t allow builders to create chatbots that pretend to be real people (e.g., candidates) or institutions (e.g., local government). 

• We don’t allow applications that deter people from participation in democratic processes—for example, misrepresenting voting processes and qualifications (e.g., when, where, or who is eligible to vote) or that discourage voting (e.g., claiming a vote is meaningless).

• With our new GPTs, users can report potential violations to us.

Better transparency around image provenance—including the ability to detect which tools were used to produce an image—can empower voters to assess an image with trust and confidence in how it was made. We’re working on several provenance efforts. We implemented the Coalition for Content Provenance and Authenticity’s⁠(opens in a new window) digital credentials—an approach that encodes details about the content’s provenance using cryptography—for images⁠(opens in a new window) generated by DALL·E 3.

We are also experimenting with a provenance classifier, a new tool for detecting images generated by DALL·E. Our internal testing has shown promising early results, even where images have been subject to common types of modifications. We plan to soon make it available to our first group of testers—including journalists, platforms, and researchers—for feedback.

Finally, ChatGPT is increasingly integrating with existing sources of information—for example, users will start to get access to real-time news reporting globally, including attribution and links. Transparency around the origin of information and balance in news sources can help voters better assess information and decide for themselves what they can trust.

In the United States, we are working with the National Association of Secretaries of State⁠(opens in a new window) (NASS), the nation's oldest nonpartisan professional organization for public officials. ChatGPT will direct users to CanIVote.org, the authoritative website on US voting information, when asked certain procedural election related questions—for example, where to vote. Lessons from this work will inform our approach in other countries and regions. 

We’ll have more to share in the coming months. We look forward to continuing to work with and learn from partners to anticipate and prevent potential abuse of our tools in the lead up to this year’s global elections.