39 Campaigns, One Banned Tool: Inside the FEC Filings That Expose ChatGPT's Quiet Role in the 2026 Midterms

The rule could not be more plainly written. OpenAI’s usage policies prohibit using ChatGPT to generate content for political advertising or targeted campaign material — a restriction the company put in place well before the 2026 election cycle and reiterated in its May 2026 election safeguards announcement. Yet a Washington Post review of Federal Election Commission disclosures, published September 5, found at least 39 US congressional campaigns that reported payments to AI companies this cycle, with consultants telling the paper many more use the tools without ever disclosing it.

It is the first systematic count of American campaigns commercially using AI since ChatGPT tightened its political-use policy — and it lands at the exact moment the AI industry is pouring a reported $265 million into the same midterm elections, making the boundary between “AI company” and “political actor” increasingly awkward to police.

What the filings actually show

The Post’s analysis of campaign finance disclosures found 39 congressional candidates who reported paying for an OpenAI subscription this election cycle. Most of those payments are mundane on their face — a $20-per-month ChatGPT Plus plan expensed under vague categories like “software” or “consulting.” But at least two campaigns disclosed that they used their OpenAI subscriptions specifically for advertising, which directly violates the company’s terms. A third candidate admitted using unspecified AI tools for drafting and personalizing messages, or for creating synthetic media.

The dollar amounts are small, which is precisely the point. Representative Mike Lawler, a New York Republican, leads the individual spending chart at roughly $3,260 in OpenAI payments since 2025. On the committee side, the Republican National Committee reported approximately $9,700 in payments to OpenAI — the largest among political action committees. These are rounding errors in a cycle where individual House races routinely burn through millions, yet they represent the visible tip of adoption that disclosure data almost certainly understates.

Campaign staffers can pay for personal subscriptions, use free tiers, or route payments through vendors and consultants whose invoices never specify the underlying tool. The 39-candidate figure captures only what was voluntarily disclosed.

Why the ban is unenforceable as written

OpenAI’s political-use prohibition is a terms-of-service rule, not a law. Enforcement relies on the company detecting prohibited use inside private conversations — and OpenAI does not station compliance officers reviewing every ChatGPT session. The Post’s reporting shows campaigns routing around the rule with little friction: a staffer drafting ad copy in a personal account, or a consultant running the model and invoicing the campaign for “marketing services,” leaves no fingerprint in FEC data.

This creates a structural asymmetry. The ban is easy to state in a policy document and near-impossible to verify in practice. And the incentive to circumvent it is growing: generative AI has become a genuine line item on campaign budgets, a leap from the anecdotal usage of 2022–2023 when ChatGPT first opened to the public.

The regulatory vacuum underneath

Federal election law has not caught up to generative AI. The FEC has been debating whether to require disclosure of AI-generated campaign content, but as of mid-2026 no binding rules have been finalized. Some states have moved faster — passing laws that require disclaimers on AI-generated political ads or restrict synthetic media depicting candidates — producing a patchwork that national campaigns can navigate by choosing where and how content is deployed.

The result is a three-way gap: OpenAI’s policy is voluntary and unenforced, the FEC’s rulemaking is stalled, and state laws vary by jurisdiction. Into that gap flows real money and real persuasive capability. The same election cycle has seen the AI industry commit roughly $265 million to super PACs and political groups, and OpenAI’s own president Greg Brockman and his wife reported $25 million in combined contributions — meaning the company whose terms prohibit campaign ad generation is simultaneously a major source of political funding through its leadership.

Why it matters beyond $3,000 subscriptions

The small dollar figures are misleading. What the FEC filings reveal is normalization: AI tools have crossed from experimental to routine in campaign operations, with disclosure lagging far behind adoption. Every undisclosed AI-drafted ad, every synthetically personalized message, every consultant invoice that launders model output through human branding chips away at the assumption that political communication is human-authored and accountable.

For OpenAI, the timing is uncomfortable. The company published its 2026 election safeguards in May, touting increased transparency and support for cyber defenders. Four months later, the first systematic look at actual campaign spending shows its own customer base includes dozens of campaigns, at least two of which openly disclosed using the product for the exact purpose the policy forbids. If the ban cannot survive contact with FEC disclosure forms — the most transparent corner of campaign finance — its prospects against undisclosed use are dim.

The deeper question the Post’s count raises is who governs AI in politics when neither the vendor nor the regulator effectively does. The 2026 midterms, now two months away, are the first national test of that question at scale — and the filings suggest the answer, so far, is nobody.

Sources

• Washington Post: “ChatGPT bans campaigns from using AI to make ads. They’re doing it anyway.” (Sept 5, 2026)
• CryptoBriefing: “39 congressional candidates are paying for OpenAI subscriptions this election cycle” (Sept 5, 2026)
• OpenAI: “Election information and safeguards in 2026” (May 27, 2026)