Rep. Luna Says Staff Used AI for Spellcheck, Not to Draft Defense Bill
Rep. Anna Paulina Luna (R-FL) is pushing back after reports surfaced that her staff used Anthropic's Claude to help draft an amendment summary for a major defense funding bill. Luna insists AI was only used for 'spellcheck' and that no legislation was ever written by AI.
Original sourceRepresentative Anna Paulina Luna (R-FL) found herself in damage-control mode this week after it emerged that her office had used Anthropic's Claude in the drafting process for an amendment summary attached to a major defense appropriations bill. Luna was quick to draw a line: the AI was used for spellcheck and light editing on a summary document, not on the legislative text itself. Her statement was emphatic — 'NO Legislation is ever written by AI' — a capitalization choice that suggested the denial was intended for a skeptical audience.
The distinction Luna draws — between amendment summaries and amendment text — is one that will likely not satisfy critics. Summary documents are themselves consequential: they shape how legislators, staff, and the public understand what a bill does. If those summaries are being quietly polished or generated by large language models, the chain of accountability in the legislative process gets murkier in ways that are hard to audit after the fact.
The incident puts a spotlight on a gap in congressional AI policy. While some offices have moved to establish internal guidelines for AI use, there is no chamber-wide standard governing when, how, or whether AI tools can be used in drafting legislative documents or their supporting materials. The defense bill context amplifies the stakes — errors or framings introduced into funding amendments for military programs carry downstream consequences that spellcheck metaphors tend to understate.
Anthropics's Claude being named specifically also matters for the broader industry. As AI companies court government contracts and position their models as trustworthy for sensitive use cases, public incidents like this — where the same tools show up in contested political contexts — complicate those narratives. Whether this accelerates congressional rulemaking on AI use or simply blows over as a one-news-cycle story may depend on whether any other offices are quietly doing the same thing and haven't said so yet.
Panel Takes
The Skeptic
Reality Check
“The 'just spellcheck' defense is the 2026 version of 'I only use ChatGPT for brainstorming' — it's a framing designed to minimize scrutiny, not provide it. The real question isn't whether AI wrote the amendment text verbatim; it's whether an LLM shaped the summary language that other legislators relied on to understand what they were voting for. Until Congress mandates disclosure logs for AI-assisted drafting, every office can claim 'spellcheck' and nobody can prove otherwise.”
The Futurist
Big Picture
“The thesis being stress-tested here is: AI assistance in government drafting is inevitable, and the question is whether it happens transparently or covertly. What this incident reveals is that the covert path is already well underway — and the accountability infrastructure to catch it doesn't exist yet. The second-order effect isn't about this one amendment; it's that the absence of disclosure norms means AI influence on legislation will scale silently until a consequential error forces the reckoning that a 'spellcheck' controversy didn't.”
The PM
Product Strategy
“The job-to-be-done here — helping overworked congressional staff produce accurate, clear legislative summaries faster — is completely real, and pretending AI isn't already doing it is the actual policy failure. The problem is there's no product decision that defined what 'appropriate AI use' looks like in a legislative office, so every staffer is making that call individually with no guardrails. What Congress needs isn't a ban; it's an opinionated internal workflow that tells staff exactly where AI is in-bounds and creates a disclosure trail — right now there's neither.”
The Founder
Business & Market
“Anthropic catching a named mention in a congressional controversy is a two-sided coin: it confirms Claude is actually being used in high-stakes government contexts, which is a market validation story, but it also means the company is now one bad headline away from being dragged into a procurement freeze or a hostile committee hearing. The government AI market is enormous, but it requires an entirely different risk surface than consumer or enterprise SaaS — and getting caught in a 'did AI write the law?' story before federal AI-use guidelines are settled is exactly the kind of reputational landmine that derails those deals.”