OpenAI Opens Election Operations Center for Nov. 3 Midterms

OpenAI told Politico it opened an Election Operations Center this week, running until every vote is counted in the Nov. 3 US midterms. Josh Lawson, the company's head of civic and strategic risk, is leading it, with "staffing in the hundreds" pulled from multiple teams across the company.

The center's job, Lawson said, is to bring "our teams together to respond quickly to emerging issues and enforce our policies as people head to the polls." That translates into monitoring real-time abuse of ChatGPT, fielding reports of unexpected model behavior, and routing voters to OpenAI's partners: Democracy Works, announced in May for registration and polling-place information, and the Associated Press for live results.

A separate safeguard, which OpenAI calls cartoonification, converts photorealistic election image prompts into cartoon output. Lawson told CNET the goal is to "degrade the evidentiary value of an image" and make ChatGPT "less useful to bad actors by adding another layer of friction." A Brennan Center report found ChatGPT was the only one of six popular AI tools making significant progress against election disinformation on that front.

That framing has a hole in it. AI Weekly readers already saw the cartoon rule break under a one-step follow-up prompt, which is the kind of workaround the new center will now be fielding at scale. It is the latest beat in our ongoing OpenAI coverage, and the one most likely to be judged on what the company discloses after Nov. 3 rather than what it is announcing today.