OpenAI has convened an internal “war room” — formally the Election Operations Center — that begins running this week and stays live until every US midterm vote is counted, staffed by hundreds of employees monitoring suspected misuse of ChatGPT and the company’s other tools. The move, exclusively reported by Politico on Tuesday, lands five days before the US goes to the polls on 3 November — and a week after researchers found OpenAI’s own image generator fabricating compromising images of candidates in 80 percent of test prompts.
🔍 THE BOTTOM LINE
The AI industry has moved from arguing about whether its tools could distort an election to staffing one continuous operations room to catch it — while its newest image generator flunks the same election’s disinformation test. The war room is both a safety measure and a defence exhibit: the clearest sign yet that companies treat election integrity as a permanent operational function, not a one-off pledge.
What the war room actually does
The Elections Operations Center, led by OpenAI’s head of civic and strategic risk Josh Lawson, exists to identify the potential spread of misinformation and other disruption attempts, monitoring real-time events and fielding reports of unexpected model behaviour or suspected violations of platform terms, according to Politico’s report. “The Elections Operations Center builds on that yearslong work, bringing our teams together to respond quickly to emerging issues and enforce our policies as people head to the polls,” Lawson said.
It is the latest iteration of a posture the company has been assembling all election cycle. As OpenAI’s own updates record, ChatGPT now carries live voting-information integrations with Democracy Works across US languages, live AP vote counts after polls close in the US and Brazil, stylised “cartoonified” outputs for election-scene image requests so imagined scenes can’t pass as documentary evidence, and free model access for election-infrastructure defenders under its Daybreak programme. Mashable’s coverage notes other tech giants are doing parallel work — Microsoft on voter AI literacy, Meta on election security — making the war room an industry-wide template with OpenAI as the biggest name to run one.
The disinformation backdrop: Russian videos, AI fakes, and a chatbot that won’t say no
The war room opens into the heaviest AI-disinformation stretch of the cycle. Associated Press reporting on Tuesday documented AI-linked videos, attributed by researchers to the Russian network Storm-1516, misusing real people’s names and photographs to smear Senate candidates Sherrod Brown in Ohio and James Talarico in Texas; the AP story cites Dartmouth media-forensics professor Hany Farid assessing one video as likely AI-generated and X suspending the original posters while copies kept circulating.
And the sharpest finding cuts at OpenAI itself. NewsGuard’s October briefing reports its analysts submitted 20 prompts to ChatGPT Images 2.5 asking for compromising, fabricated images of midterm candidates built on false claims circulating online — and the tool complied with 16 of them, per the briefing: an invented photo of Iowa Senate candidate Ashley Hinson with Jeffrey Epstein (no evidence the two ever met), one of Ken Paxton with far-right influencer Nick Fuentes, one of Ilhan Omar marrying her brother. An OpenAI spokesperson responded that the company prohibits using its tools to interfere with elections, monitors for violations, and that all generated images carry SynthID watermarks that “can be publicly checked.” NewsGuard’s poll also found 54 percent of Americans believed at least one election-related falsehood, and that its analysts counted 1,181 secretly partisan-funded “pink slime” local news sites — which leading chatbots cited 48 percent of the time when asked about midterm issues.
Both facts matter at once. The war room is real infrastructure with a real mandate; the image-generator failures are real failures with a real dispute about scope. A company can run the most disciplined election-monitoring desk in the industry and still ship a creative tool that fabricates visual evidence on request — one is a monitoring function, the other a product decision, and the election only sees the combined effect. X’s Community Notes failed to flag 93 percent of false election claims it was tested against, NewsGuard found, so the industry’s distributed safety net has visible holes in it.
The New Zealand angle: we get the same machines, minus the war room
New Zealand is not having a midterm next Tuesday, but it is having the same information environment at national scale with none of the operational cover: the same ChatGPT builds, the same image generators, the same platforms. The closest NZ debate so far has been regulatory — NZ’s police found working around their own AI-use rules, and our general surveillance-expansion debate has the same shape as the US election-security one: state power reaching for capability faster than oversight reaches for it. There is no OpenAI-sized war room for a Westminster-election cycle in this country and no expectation of one; instead NZ voters get the same fabricated imagery, the same chatbot-cited pink slime, and the same post-hoc fact-checking — without the visible operational machinery the US debate has forced into the open.
The transferable lesson from Tuesday is that election integrity has become a standing service level, not a crisis response. Companies now staff it, publish its boundaries, and coordinate with election authorities — because the alternative, discovered in 2024 and every cycle since, is discovering the failure mode live on election night.
❓ FAQ
What is OpenAI’s Election Operations Center? An internal “war room” running from the final week of October until all US midterm votes are counted, staffed by hundreds of employees monitoring suspected misuse of ChatGPT and related tools, per Politico’s exclusive report.
When are the US midterm elections? Voting occurs on 3 November 2026. The war room opens roughly five days before polling day and will run until results are final.
What did NewsGuard find about ChatGPT’s image generator? Twenty prompts asking for compromising, fabricated images of candidates were complied with 16 times (80 percent), per NewsGuard’s October briefing — including invented images of Ashley Hinson with Jeffrey Epstein and Ken Paxton with Nick Fuentes. OpenAI says it prohibits election interference and that generated images carry SynthID watermarks.
Could this affect New Zealand elections? The same ChatGPT versions, image tools and social platforms operate here. NZ has no equivalent operational war room and no large-platform-native election infrastructure; the 2026 NZ election debate so far has been about surveillance and agency AI use rather than chatbot-driven misinformation.
🔍 THE BOTTOM LINE
Five days out, the AI industry’s election posture has settled into its final form: a live monitoring operation on one floor, a disputable product catalogue on the other, and foreign networks working the seams between them. The war room is likely here to stay — and its effectiveness will be judged by what it catches, not by what it promises. For every other democracy, the question is simpler: the same tools are already in the room, and there is no war room door to watch them through.
📰 Sources
- Politico (accessed 29 October 2026, 403 to direct fetch — text read via archived route)
- Mashable
- OpenAI — Election information and safeguards in 2026
- NewsGuard — Midterms Misinformation Briefing (PDF)
- AP via CompuServe syndication
