OpenAI has now laid out in detail how it is approaching the problem as voters head to the polls across the United States and around the world.

Elections have always been a target for disinformation, but the rise of generative AI has made the threat far more sophisticated and harder to detect.

OpenAI Adds Election AI Deepfake Protections

OpenAI published an updated safeguards page for 2026 detailing the specific measures the company is putting in place to protect elections from AI-driven interference, with a particular update added on October 20 ahead of the US midterm elections. The page outlines five primary focus areas: surfacing reliable voting information, supporting cyber defenders, increasing transparency around AI-generated content, combatting misuse, and monitoring political bias in its models.

Jesse Paul on Unsplash

One of the most significant additions is a partnership with Democracy Works, a nonpartisan organization that compiles election information directly with state and local election officials. The integration is now live in the United States across multiple languages inside ChatGPT, giving users direct access to registration deadlines, voting locations, and other election logistics.

Responses also include links to official election sources so users can verify the information independently.

OpenAI is also bringing live vote count data into ChatGPT through a partnership with The Associated Press, covering elections in the United States and Brazil as results come in on election night. The company made clear that election officials remain the authority on certified results, with the AP integration providing real-time data rather than analysis or projection.

OpenAI Tightens Defenses Against AI Deepfakes

The transparency measures OpenAI has deployed ahead of this election cycle directly target the deepfake problem. The company announced a partnership with Google DeepMind's SynthID to embed invisible digital watermarks into images generated through ChatGPT, the OpenAI API, and its Codex platform.

That watermarking layer is designed to survive common transformations like screenshots and cropping, maintaining its signal even after an image has been passed around online.

The SynthID watermarking works alongside OpenAI's existing use of the C2PA standard, a cryptographic metadata system that attaches provenance information directly to image files. C2PA provides richer contextual data than a watermark alone, and the two systems working together give platforms and users multiple signals to check whether content originated from an AI system.

OpenAI has also launched a public verification tool at openai.com/verify that allows anyone to check whether an image they encounter on the internet was generated using OpenAI tools. The tool detects the presence of a SynthID watermark and surfaces any C2PA metadata attached to the file, giving journalists, election officials, and ordinary users a direct way to investigate content they are unsure about.

On the image generation side, the company now applies a stylization layer to requests for images depicting election interference or content that could discourage people from voting. Instead of producing photorealistic images that could pass as documentary evidence, the system generates cartoonified illustrations for those requests, reducing the risk that an AI-fabricated scene is mistaken for a real photograph.

The company noted that its policies prohibiting election interference and voter demobilization still apply even when the resulting image is stylized.

Fraudulent AI Activity

The misuse enforcement side of OpenAI's safeguards covers both automated and human-directed attempts to exploit its tools for election interference. OpenAI's Intelligence and Investigations, Safety, Security, and Integrity teams actively monitor for patterns of misuse and investigate attempts to deploy its tools in deceptive or covert influence operations.

Since February 2024, OpenAI has published regular reports documenting its findings on disrupted influence operations, including state-affiliated activity, and it confirmed it will continue publishing those reports through this election cycle. When the company identifies abuse, it takes enforcement action that can include restricting or permanently terminating access to its services.

Political advertising is also blocked outright. OpenAI continues to prohibit advertisers from running political ads on its platform and bars users from creating or distributing scaled campaign messaging, including AI-generated images and video, that advocates for or against any candidate, political party, or ballot measure. Campaigns are permitted to use OpenAI tools for internal work such as briefings, translation, planning, compliance, and administrative tasks, but any use that crosses into mass political persuasion is explicitly prohibited.

On the cybersecurity side, OpenAI has extended access to its Daybreak program to voting system manufacturers registered with the US Election Assistance Commission and to state election authorities. Daybreak includes Codex Security, which automatically identifies and helps remediate vulnerabilities in code, and a Trusted Access for Cyber program that gives verified cyber defenders access to frontier models.

OpenAI is also working directly with the National Association of Secretaries of State and the National Association of State Election Directors to brief election officials on the latest cyber tools available to them.

Frequently Asked Questions

What Is an AI Deepfake?

An AI deepfake is synthetic media, whether video, audio, or imagery, generated or manipulated by artificial intelligence to portray a person saying or doing something they never actually said or did. The term comes from "deep learning," the AI technique underlying the most capable generation and manipulation tools, combined with "fake." Early deepfakes required significant technical skill and computing power to produce, but modern generative AI tools have made high-quality synthetic media accessible to anyone with an internet connection.

The result is content that can be convincingly realistic and difficult to distinguish from authentic recordings without technical analysis. Deepfakes are used across a wide range of contexts, from harmless creative applications to malicious ones including fraud, nonconsensual intimate imagery, and, increasingly, political disinformation designed to spread false information about candidates or election processes.

How Can Deepfakes Disrupt Elections?

Deepfakes can interfere with elections in several distinct ways. The most direct is voter suppression or demobilization: a fabricated audio clip or video of an official announcing that an election has been postponed, that polling locations have changed, or that a candidate has withdrawn can discourage people from showing up to vote.

A deepfake of a candidate making an inflammatory or false statement can shift public opinion in the final days before an election when there is limited time for fact-checking to reach voters at scale. Deepfakes can also be used to make it appear that a candidate said or endorsed something they did not, undermining trust in the integrity of the electoral process even when the fabrication is eventually debunked. At a systemic level, the widespread availability of convincing synthetic media erodes the baseline public trust in authentic video and audio as evidence, which benefits anyone who wants to dispute genuine footage as fake.

What Defenses Has OpenAI Prepared?

OpenAI's election defenses in 2026 operate across five layers. The first is information reliability, handled through partnerships with Democracy Works for voting logistics and The Associated Press for live vote counts, both integrated directly into ChatGPT. The second is content provenance, addressed through SynthID watermarking and C2PA metadata on all AI-generated images, along with a public verification tool at openai.com/verify.

The third is image generation guardrails, which convert requests for photorealistic election interference imagery into stylized illustrations rather than realistic photos.

The fourth is active enforcement through its Intelligence and Investigations team, which monitors for misuse patterns, publishes regular reports on covert influence operations it has disrupted, and terminates access for accounts found to be interfering with elections.

The fifth is political neutrality monitoring, through a formal bias evaluation process built into how OpenAI tests its models to ensure ChatGPT does not nudge users toward or away from political positions when they engage with civic questions.