OpenAI election war room — hundreds of staff watch ChatGPT misuse through the midterm vote count

An AI company has set up an election war room, as social networks have done for years. OpenAI has opened an Election Operations Center ahead of the November 3, 2026 US midterm elections, Politico reported on October 28. Hundreds of employees from several teams are involved, and it will run until the votes are counted.

What it watches is OpenAI's own products, not the wider internet: whether ChatGPT and other tools are being used to interfere with the election, whether models behave unexpectedly, and whether accounts are breaking the rules. Platforms such as Meta have run similar centers in past election cycles; AI companies rarely have.

1. What the center does

Lawson told Politico the center brings teams together to respond quickly to emerging issues and enforce policies as people go to vote. He described it as an extension of years of work steering users to reliable voting information and supporting election officials.

For readers outside the US: the midterms elect all 435 House seats and about a third of the Senate, two years into a presidential term. Polls close on the evening of November 3 US time, which is the morning of November 4 in Korea.

2. Why now — two operations exposed in October

Three weeks before the center opened, OpenAI's October 8 report said it had banned accounts tied to two covert influence operations from Russia and Iran. Politico said the center will watch for similar activity.

OpenAI says it has exposed 30 covert influence operations in the past two and a half years. Dark Clark's operators mostly used ChatGPT to write internal reports for their superiors; generating fake content was a smaller part of the work. OpenAI said it shared findings on both with authorities.

3. The safeguards running alongside

Democracy Works data covers federal, state, county, municipal and school-board elections in jurisdictions with more than 5,000 residents, including Election Day polling-place changes and extended hours, sourced directly from election officials.

The cartoon rule is new for this cycle. According to French tech site Clubic, a request showing enforcement agents blocking a polling station comes back as a drawing even when the user insists on realism. Lawson told CNET the goal is to degrade the evidentiary value of such images and add another layer of friction for bad actors; the problem with deepfakes, he argued, is deception rather than social commentary. In Clubic's tests, though, a stuffed-ballot-box scene first came out as a drawing and then as a photorealistic image on a second request.

4. AI's new place in elections

As more voters ask chatbots about voting, AI companies carry responsibility not as platforms that spread content but as the source of the answer itself. A recent study found that through developer APIs Grok picked Republican candidates in 92% of races and GPT picked Democrats in 97%, while consumer apps declined the same question. Politico noted that some 2026 campaigns have made criticism of AI a central message, so a single incident could quickly become a political issue for OpenAI.

5. What remains unknown

• Headcount and authority: Beyond "hundreds," who staffs it full-time, who can ban accounts and how it coordinates with outside agencies are not public.
• Post-election disclosure: Whether OpenAI will publish the number of incidents or accounts banned is not known.
• Cartoon rule after the election: OpenAI has not said whether it will keep the rule.
• Source concentration: Most details about the center come from a single Politico report.