US campaigns' AI spend: what FEC data shows, and the data risks it raises
New Federal Election Commission disclosures show US political campaigns buying AI tools, mostly for back-office work rather than deepfakes. For security teams, the real question is where sensitive campaign and voter data goes once it enters these tools.
Key Takeaways
- An analysis of FEC itemized expenditures, written by Bruce Schneier and Nathan E. Sanders, finds at least $17 million in disclosed AI spending across 523 federal candidates and committees since 2020.
- Spending on general-purpose LLM providers is small but growing: roughly $50,000 to OpenAI since 2024, and more than 65 candidates and committees paying Anthropic in 2026.
- Synthetic-media tools such as ElevenLabs and Midjourney show minimal disclosed spend. The visible adoption is in everyday workflow tools, not deepfake generation.
- Campaigns are small, short-lived organisations that handle sensitive data. Putting that data into third-party AI services is a governance and third-party-risk problem.
A new analysis by Bruce Schneier and Nathan E. Sanders, published on Schneier on Security and originally in The Guardian, uses itemized expenditure data from the US Federal Election Commission to show how campaigns are paying for AI. The authors note that candidates themselves are generally quiet about how they use the technology. The spending records are the more reliable signal.
What the disclosures show
According to the essay, at least $17 million in AI-related spending has been disclosed across 523 federal candidates and committees since 2020. At state level, the authors found more than $92,000 across California, Colorado, Massachusetts and Washington since 2022.
- OpenAI: about $50,000 across more than 80 federal campaigns and committees since 2024. The Republican National Committee is the largest buyer, at around $10,000.
- Anthropic: more than 65 candidates and committees in 2026, up from essentially zero in earlier cycles. Senator Tom Cotton's campaign spent over $4,000.
- xAI (Grok): about $5,000 across nine federal and state entities, with RFK Jr.'s 2024 campaign the main user.
- Campaign-specific platforms: Daisychain at $300,000 in 2026, up from $50,000 in 2024. Prompt.io at $375,000 in 2026, down from $500,000 in 2022. AmplifAI at $4.7 million in the 2022 cycle alone.
Deepfake tools barely register
Public debate tends to centre on synthetic media. The disclosed spending points elsewhere. The essay reports six federal users of ElevenLabs totalling $1,400, and five of Midjourney totalling $1,600. Separately, it cites Wesleyan Media Project data showing at least 164 political ads containing AI-generated media, backed by $80 million in ad spending. It also cites a September 2025 Pew survey in which more than 70% of respondents would view candidates negatively for AI-generated speech assistance. That helps explain why campaigns say little about their use.
Why this matters to security teams
The following is our analysis, not a finding from the essay. FEC data records payments, not what staff paste into the tools. Still, a campaign is a good example of a high-risk AI adopter: a temporary organisation with a small or volunteer IT function, a high-value target, and a steady flow of donor, voter and strategy data.
- Shadow AI and data egress. Drafts, donor lists and internal strategy can flow into consumer or lightly governed AI accounts with no retention, access or audit controls.
- Third-party concentration. Campaign-specific vendors hold aggregated data from many clients. A breach at one is a breach for all of them.
- Prompt injection and agent access. Tools that read email, documents or the web inherit injection risk. Any assistant with write or send permissions needs the same scrutiny as any other privileged integration.
- Disclosure opacity. Because the technology is used quietly, there is little public visibility into what controls exist.
Practical steps for any small, high-risk organisation
- 1Publish an approved-tools list and block or monitor unsanctioned AI services.
- 2Classify what may never be entered into a third-party model, such as voter files, donor financials and unreleased strategy.
- 3Review each AI vendor's retention, training-use and sub-processor terms before signing.
- 4Give AI assistants least-privilege access and keep a human approval step on outbound actions.
- 5Test assistants that touch email and documents for prompt injection before launch.
Frequently Asked Questions
Where does the data on campaign AI spending come from?
The essay draws on itemized expenditure disclosures filed with the US Federal Election Commission, supplemented at state level by campaign finance records from four states. These records show payments to vendors, not how the tools are used.
Are campaigns mainly spending on deepfake tools?
Not according to the disclosed data. The essay reports only $1,400 on ElevenLabs across six federal users and $1,600 on Midjourney across five. Larger sums go to campaign-specific platforms and general-purpose LLM providers.
What is the main security risk of campaigns using AI tools?
Our assessment is that it is uncontrolled data handling. Sensitive donor, voter and strategy material may reach third-party services without clear retention limits, access controls or vendor due diligence.
Sources
- 1How American Political Campaigns Are Using AI—and What They're Spending on the Tools — Schneier on Security
- 2Campaign finance data (itemized disbursements) — Federal Election Commission
- 3Wesleyan Media Project — Wesleyan University