OpenAI and Anthropic Run Their First Election War Rooms as Americans Vote in Midterms

The teams are there to catch misuse of the companies' AI products, to deal with unexpected model behaviour and to make sure users get accurate voting information. The New York Times reported on Tuesday that OpenAI, Anthropic and other technology companies have pledged to protect the vote from efforts to undermine democracy.

What OpenAI Has Set Up

Politico first reported OpenAI's effort, which the company calls its Election Operations Center. Josh Lawson, OpenAI's head of civic and strategic risk, leads it. According to accounts of that report, the center draws staff in the hundreds from several teams across the company. They watch for real-time abuse of ChatGPT, handle reports of unexpected model behaviour and investigate suspected breaches of OpenAI's usage terms. The center opened in election week and will stay active until the votes are counted.

The center deals only with OpenAI's own products, not with election misinformation elsewhere online. OpenAI also has two partners for election information inside ChatGPT. Democracy Works, a nonpartisan group, supplies voter registration and polling-place details, and the Associated Press supplies live results after polls close. For election-related prompts that ask for photorealistic images, OpenAI turns the output into a cartoon-style picture. Lawson told Politico the aim was to "degrade the evidentiary value of an image."

The work follows OpenAI's disclosure in October that influence operations linked to Russia and Iran had used ChatGPT to help produce deceptive political material.

Anthropic's Approach

An Anthropic spokesperson told Semafor that the company's teams work on election safeguards all year. "Ahead of the election, we have a dedicated war room monitoring our platform for misuse," the spokesperson said, adding that the team was ready to respond quickly to any incidents. Reports say Claude shows users who ask about voting a banner that links to TurboVote, a Democracy Works service.

The idea is not new in technology. Meta set up its first election war room in 2018 because of concerns about misinformation on its platforms. Katie Harbath, a former Meta director of public policy, told Semafor that such rooms are "built for dealing with the things we can't necessarily predict going wrong," and that they need a fast, settled triage process.

Why It Matters

This is the first big US election since conversational AI went mainstream, and the labs' tools now sit between many voters and basic election information. Semafor noted that last week Google's AI Mode failed to answer a question about who won the 2020 election, and that Google said it was working on a fix. Separately, researchers have found frequent factual errors in chatbot answers about the midterms.

The war rooms are voluntary measures that the companies run themselves. There is still no federal law covering deceptive AI in elections. Executives and policymakers will be watching how quickly the labs spot and fix problems on election day, and whether they publish an account of what they found once counting ends.