OpenAI and Anthropic have set up dedicated election “war rooms” to watch for misuse of their AI tools ahead of Tuesday’s midterm elections, Nextgov/FCW reported Monday, citing Politico and Semafor. The move puts AI developers alongside social media companies in election-integrity work, even as an independent study found that safeguards against fake election images are uneven across the industry.
What Happened
According to Nextgov/FCW, the war rooms are meant to give users reliable voting information and curb misinformation, particularly misleading AI-generated images. Neither OpenAI nor Anthropic responded to the outlet’s requests for comment.
Other companies described their own steps. A Meta spokesperson pointed Nextgov/FCW to a company post saying Meta activated its Election Operations Center, a command center it says it has used for dozens of elections worldwide, in the week before Election Day. Microsoft said it has improved how its Copilot assistant answers election questions, is still testing those answers and directs voters to state and local election officials. The company said malicious deepfakes, realistic fabricated images, audio or video, of candidates and election officials are not allowed on its platforms. Google and Perplexity did not respond, the outlet said.
Why It Matters
Voluntary safeguards vary widely. The Brennan Center for Justice at NYU School of Law, a research institute, reran earlier prompts against six image generators for its Oct. 22 report, “Major AI Models Are More Willing than Ever to Create Fake Election Images.” It found that OpenAI’s ChatGPT had significantly improved since spring and now steers potentially harmful election prompts into a cartoon-like style unlikely to pass as real photos. Google’s Gemini and Meta AI became more willing to produce convincing fakes. Runway and xAI’s Grok changed little, and results for Black Forest Labs’ Flux were mixed. The authors called on legislators, regulators and state attorneys general to set rules and liability for misuse.
For boards and executives, the lesson reaches beyond AI labs. Any organization that puts a chatbot in front of the public on a sensitive topic needs monitoring, an escalation path and the ability to correct answers fast. Attackers are already using AI against the people who run elections: the FBI and the Cybersecurity and Infrastructure Security Agency (CISA) warned officials of AI-polished phishing in October, as The Primary Wire reported.
Lawmakers’ Pressure
Sen. Mark Warner, D-Va., told Nextgov/FCW it was promising that leading AI companies are working to give accurate election information and “immediately detect and interrupt abuse attempts.” In October, Warner, Sen. Alex Padilla and Reps. Jim Himes and Joe Morelle, all Democrats, asked companies to base AI answers on election questions only on “highly authoritative sources” rather than social media. The Primary Wire reported on that letter, citing Nextgov/FCW.
What to Watch
Whether companies or election officials disclose incidents involving AI tools during and after the vote, and whether the companies publish what their war rooms found. Those accounts will shape the debate in Congress over whether voluntary safeguards are enough.
