OpenAI opens an election war room for the midterms

OpenAI opens an Election Operations Center before the midterms

OpenAI is running an election "war room" for the midterms, which it calls the Election Operations Center. The company gave the details exclusively to Decoded, Politico's technology newsletter, and Politico published the story on October 28. The midterms are the US elections for Congress held halfway through a president's term, and this year's vote is on November 3.

It watches for misuse of ChatGPT until every vote is counted

The center opened the week of October 28 and will keep running until all votes are counted. Its job is to spot misinformation and other attempts to disrupt the vote, with most of the attention on people abusing ChatGPT and other OpenAI tools. According to Politico, the team will:

• follow events as they happen
• take reports of AI models behaving in ways nobody expected
• take reports of possible breaches of OpenAI's terms and review cases flagged as potential abuse

It will also keep an eye on two features OpenAI announced in May. A partnership with the nonpartisan group Democracy Works gives voters registration and polling place details inside ChatGPT, and a deal with the Associated Press feeds in live election results.

Hundreds of staff from several teams, led by Josh Lawson

Staff come from multiple teams across OpenAI, and the headcount runs into the hundreds. Josh Lawson, OpenAI's head of civic and strategic risk, leads the effort.

Lawson told Politico the company has spent years on reliable voting information, on stopping misuse of its tools and on supporting the people who defend elections. He described the center as pulling those teams into one place so they can react fast and enforce OpenAI's rules while people vote.

The center follows OpenAI's report on Russian and Iranian operations

The center arrives the same month OpenAI disclosed two influence operations that ran on its models. An influence operation is a campaign that hides who is behind it while trying to sway public opinion.

Two false front operations used OpenAI models in October

Earlier in October, OpenAI reported that two operations run by Russia and Iran had used its models. Both hid behind invented organizations or people, a setup Politico calls a "false front," and pushed messages about geopolitics and conflict to the audiences they targeted. Our earlier report, OpenAI bans Russian, Iranian ops using fake fronts, covers how each one worked.

Worries about AI images, audio and video ahead of the vote

Politico describes a wider fear that AI could skew the midterms. It points to two ways that could happen:

• misleading images, audio and video generated with AI
• misinformation spreading online

A deepfake is a realistic fake image, video or voice made with AI. Some campaigns have also made opposition to AI a rallying point, which puts OpenAI in the middle of the fight. Politico expects the pressure on the company to grow as its tools reach further into Americans' daily lives.

For an AI lab, a dedicated election center is new. Other tech companies have done it before, Meta among them.

ChatGPT answers fake polling place requests with cartoons

OpenAI builds safeguards into its models to keep people from producing misleading content. One example Politico gives is a rule that swaps a photo-like image for a cartoon.

Cartoonification keeps fake scenes from looking like photos

Ask OpenAI's image tools for a long line at a polling place that does not exist, and you get a cartoon instead of something that looks like a photo. Politico calls this "cartoonification."

The Brennan Center, a US law and policy institute, published a report the week before Politico's story. Of the six most popular AI tools, it found that only ChatGPT had made significant progress against election disinformation since May. Politico credits the cartoon rule for that progress.

The center covers OpenAI tools, not other companies' AI

The center exists mainly to stop abuse of OpenAI's own products. Images or text made with another company's AI never pass through OpenAI's safeguards, so the cartoon rule cannot touch them.

Politico notes that AI models have improved sharply since 2024 and have also escaped their makers' control several times. People who want to disrupt an election can turn to other AI systems to try to get around safeguards.

How workplaces can prepare for election deepfakes

Deepfakes are not only an election problem. In a December 2024 public service announcement, the FBI warned that criminals use AI-generated video to pose as company executives in live video calls. If someone fakes your CEO during election season or any other time, the company should already know how it will respond.

Companies should plan for deepfakes of their executives

Decide in advance who checks whether a video or image of your leaders is real, and through which channel the company will respond. The FBI's advice for suspicious calls applies here too: hang up, look up the organization's contact details yourself and call back directly. Set the same rule for any urgent request that arrives by video.

Staff should check images and AI answers before sharing them

Before you share an election image, video or AI answer, check where it came from. The FBI lists signs of a fake to look for, including distorted hands or feet, irregular faces, shadows that fall the wrong way and unnatural movement. Get voting dates and polling places from your state or local election office, not from a screenshot.

To check whether an election image or deepfake photo is AI-generated, run it through an AI image detector for a second opinion. Treat the result as one signal rather than proof, and keep checking where the image came from.