In a small conference room off the cafeteria at one of OpenAI’s offices in San Francisco, a team of more than a dozen people are working around the clock to prevent anyone from using ChatGPT to disrupt this year’s midterm elections.
The operations center represents a vast expansion of the company’s nascent efforts ahead of the 2024 presidential election. Artificial intelligence tools were far less powerful at the time and the chatbot had a fifth of the one billion users it has today.
“The most available technology at the time in 2024 was text,” said Josh Lawson, the head of civic and strategic risk at OpenAI, who is leading the operation center. “We’re not just monitoring text anymore.”
Now, the potential threats include deep fakes, cyberattacks and foreign and domestic influence operations that can be created and launched with a series of prompts.
The company is not alone. Anthropic, which operates the Claude chatbot, has set up its own “war room” for the first time, assembling teams of analysts in cells in San Francisco, New York and Washington. Google and Meta have also pledged robust efforts to thwart efforts to use their A.I. tools to undermine confidence in the democratic process.
As A.I. becomes increasingly integrated into social and political life online, the companies have all faced criticism for not doing enough to limit malicious content.
The Brennan Center for Justice, a voting rights and democracy group at New York University School of Law, found that the most popular chatbots generally pushed back on queries about election-related conspiracy theories. At the same time, the center found, A.I. tools regularly created text, images and videos that could be used to undermine confidence in elections.
The companies’ establishment of war rooms recalls Meta’s creation of one for the 2018 midterm elections following the furor over a Russian interference campaign on Facebook during the 2016 presidential election.
Since then, Meta and other social media platforms scaled back their trust and safety teams, including efforts to police election-related disinformation. They did so under pressure from Republicans and, after 2025, the Trump administration, who accused them of censoring conservative views.
That retreat has coincided with the proliferation of A.I., increasing fears of its impact on elections. Experts and former government officials have warned that technology has created a potent new ways to undermine public confidence in the electoral process even as the Trump administration dismantled the government teams that previously combated efforts to disrupt voting.
“I think it would be foolish to think that threat has dissipated and that technology is not making the threats more potent and believable,” said Lauren Protentis, a former government official who oversaw foreign malign influence on the National Security Council under former President Joseph R. Biden Jr.
OpenAI stood up its election center last Thursday, assembling a team of employees who are dedicated to monitoring misuse of the world’s most popular chatbot. The team in San Francisco works with a satellite center in Washington and involves scores of people, company officials said.
They examine prompts on ChatGPT that have been automatically flagged for attempts to create false texts or images about the election. That could include prompting the chatbot to design an influence campaign — about stuffing ballots, for example — to spread on a social media platform using fake accounts. The company notifies the platforms when it identifies malign activity intended for use on their site.
OpenAI has already disclosed efforts by Russia, China and Iran to use ChatGPT to carry out foreign influence operations, including two campaigns it disrupted last month. It has not yet disclosed any specific efforts in the midterm elections.
“We’ve been trying to catch operations,” OpenAI’s chief investigator, Ben Nimmo, said in an interview, “before they have a chance to gain traction.”
The post Tech Giants Try to Stem Flow of Election Disinfo Unleashed by A.I. appeared first on New York Times.
