A Florida congresswoman is defending her office's use of Anthropic's Claude AI after a leaked amendment summary for the 2027 National Defense Authorization Act revealed telltale signs of AI assistance. Rep. Anna Paulina Luna insists her staff only used the AI tool for "spellcheck," but the incident has ignited a broader debate about transparency and AI's role in crafting legislation that governs military spending and national security.
A seemingly routine defense spending amendment just became the latest flashpoint in Washington's struggle with artificial intelligence transparency. Rep. Anna Paulina Luna (R-FL) found herself defending her office's use of Anthropic's Claude AI after eagle-eyed observers on X spotted something unusual in an amendment summary for the 2027 National Defense Authorization Act.
The leaked document contains a smoking gun that's hard to miss. Midway through describing an amendment about border defense operations, the text reads: "Identical to H.R. 100 (118th Congress).11:25 AM????Claude responded: Requires the Secretary of Defense to designate Department of Defense activities, support, and operations at the southwest land border as a named operation with…" The phrase "Claude responded" - a clear artifact from Anthropic's AI assistant - was apparently left in the final text that circulated among congressional offices.
Luna's response came swiftly but raised more questions than it answered. In a statement, she insisted that her staff used AI only for "spellcheck" purposes when drafting the amendment summary, emphasizing in all caps that "NO Legislation is ever drafted with AI." But the distinction between using AI for proofreading versus drafting may be lost on critics who see the incident as evidence that artificial intelligence has already embedded itself deeper into the legislative process than most Americans realize.
The controversy cuts to the heart of an emerging tension on Capitol Hill. As AI tools become more sophisticated and accessible, congressional staffers - often overworked and underpaid - face mounting pressure to process massive volumes of legislative text, policy briefs, and constituent correspondence. Claude, which Anthropic markets as a safer, more transparent AI assistant compared to competitors, has gained traction in professional settings precisely because it can quickly summarize complex documents and catch errors.
But Luna's office made a critical mistake that AI experts have been warning about: they failed to properly review and clean the AI-generated output before it left their office. The presence of "Claude responded" in a congressional document is the equivalent of leaving "Comment from editor: fix this part" in a published news article. It's not just embarrassing - it reveals a workflow where AI outputs weren't adequately scrutinized by human eyes.
The timing couldn't be more awkward. The 2027 NDAA is one of the most significant pieces of legislation Congress passes each year, authorizing hundreds of billions in military spending and setting defense policy priorities. That an amendment related to border security operations - itself a politically charged topic - would become the vehicle for this AI disclosure makes the story even more combustible in an election year.
Congress has been grappling with how to regulate AI in the private sector while remaining largely silent on its own internal use of these tools. There are currently no formal disclosure requirements when congressional offices use AI to draft legislation, amendments, or even constituent communications. Some transparency advocates argue that voters have a right to know when algorithms rather than human judgment are shaping the laws that govern them, even if that AI use is limited to editing or summarization.
Anthropic hasn't commented on the Luna incident, but the company has positioned Claude as particularly well-suited for high-stakes professional work where accuracy and transparency matter. The startup, which has raised billions from investors including Google and Spark Capital, emphasizes "constitutional AI" - an approach designed to make the system's reasoning more interpretable and aligned with human values.
Yet this episode demonstrates a fundamental challenge: even the most carefully designed AI systems are only as reliable as the humans using them. Luna's staff may have indeed used Claude only for spellchecking, as claimed. But by failing to remove the AI's metadata from the final document, they've handed ammunition to critics who worry about AI's opaque influence on governance.
The fallout is already rippling through Washington. Several congressional observers have called for new rules requiring disclosure when AI tools are used in legislative drafting, similar to transparency requirements around lobbying. Others argue the incident is overblown - that using AI for editing is no different than using Microsoft Word's grammar checker, which has been quietly suggesting changes to congressional documents for decades.
What makes this controversy different is the level of sophistication. Traditional spellcheck flags typos; large language models like Claude can restructure entire paragraphs, suggest policy language, and even generate arguments for or against specific provisions. The line between "assistance" and "authorship" gets blurrier with each new model release.
Luna's Claude controversy may seem like a minor embarrassment, but it's actually a preview of much bigger battles ahead. As AI tools become standard equipment for knowledge workers across every industry, Congress will need to decide whether and how to disclose their use in lawmaking. The alternative is a steady drip of incidents like this one, where accidental transparency forces conversations that should have happened proactively. For now, the lesson is simple: if you're going to use AI to help write America's laws, at least remember to delete the part where the AI introduces itself.
