AI in Congress? Congresswoman Denies Staff Used Claude to Draft Defense Amendment
A controversy erupted in Washington when screenshots circulating on X (formerly Twitter) appeared to show text from an AI chatbot — specifically Anthropic's Claude — embedded inside an official congressional amendment summary. The document in question was tied to the 2027 National Defense Authorization Act (NDAA), one of the most significant pieces of defense legislation Congress produces each year. The screenshots raised immediate questions about whether artificial intelligence was being used to write actual U.S. legislation, and the congresswoman at the center of it all moved quickly to push back.
What the Screenshots Actually Showed
The amendment summary that sparked the firestorm contained a highly unusual passage. Within the text of the document, the following line appeared: "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…"
To anyone familiar with AI chatbot interfaces, the phrasing "Claude responded:" is immediately recognizable as output from Anthropic's Claude — a large language model AI assistant similar in function to ChatGPT. The appearance of this kind of text inside a formal legislative document set off alarm bells across social media and among those who follow congressional transparency and technology policy.
The amendment in question was submitted by Rep. Anna Paulina Luna, a Republican congresswoman representing Florida's 13th congressional district. The bill deals with defense operations at the southwest land border, a politically charged subject in the current legislative landscape.
Rep. Anna Paulina Luna's Response
Facing mounting pressure online, Rep. Luna moved swiftly to address the controversy. She acknowledged that her staff had used AI — but drew a firm line on how it was used. According to Luna, artificial intelligence was employed only for "spellcheck" purposes when drafting the amendment summary, not for the actual legislative text itself.
In her statement, Luna was unequivocal: "NO Legislation is ever drafted with AI." She insisted that the appearance of Claude's output in the document was a result of using the tool in an assistive, editorial capacity rather than as a ghostwriter for policy language.
Her clarification did little to immediately silence critics, however. Many observers pointed out that even if Claude was used only for proofreading or summarization tasks, the fact that raw AI output — including metadata-like chatbot prompts — ended up in an official government document is itself a significant oversight. The question of internal review processes and document handling quickly became part of the conversation.
Why This Matters: AI and Legislative Integrity
The incident touches on a broader, increasingly urgent debate about the role of artificial intelligence in government. As AI tools become more sophisticated and more widely available, their use in professional and governmental settings is expanding rapidly. But the guardrails around that use — particularly in high-stakes contexts like federal legislation — remain poorly defined.
There are several reasons why the use of AI in congressional drafting raises legitimate concerns:
- Accuracy and accountability: AI language models can produce plausible-sounding text that contains factual errors or ambiguous legal language. Legislation must be precise, and errors can have real-world consequences.
- Transparency: The public and other legislators have a right to know how bills are written. If AI is involved in any stage of the drafting process, that raises questions about disclosure.
- Security: Inputting sensitive legislative content into a commercial AI platform may expose confidential policy details to third-party systems, depending on the tool's data handling policies.
- Attribution: When a lawmaker's name is attached to legislation, there is an implicit expectation that their office — staffed by accountable human beings — is responsible for the content.
The Broader Trend: AI Is Already in the Building
Rep. Luna's situation is unlikely to be an isolated case. Across Capitol Hill, congressional staffers are under enormous pressure to produce high volumes of work with limited resources. AI tools offer an attractive solution for tasks like summarizing long documents, drafting routine communications, and proofreading text. It would be naive to assume that Luna's office is the only one experimenting with these tools.
Several think tanks and government transparency advocates have called for clearer guidance from congressional leadership on how and when AI can be used in the legislative process. Some members of Congress have introduced legislation aimed at regulating AI in federal agencies, though comprehensive rules specifically governing its use on Capitol Hill have yet to materialize.
The European Union has already moved toward requiring transparency around AI-generated government content as part of its broader AI Act framework. The United States, by contrast, has taken a more fragmented approach, with individual agencies setting their own informal guidelines.
Public Reaction and Political Fallout
On social media, reactions ranged from outrage to mockery. Critics of Rep. Luna used the incident to question her office's competence and to make broader arguments about the risks of AI in governance. Supporters, meanwhile, argued that the backlash was overblown — pointing out that using AI for spellchecking or editing is no different from using Grammarly or Microsoft Word's built-in editing tools.
That argument has some merit on the surface, but it misses a crucial distinction: tools like Grammarly operate locally or with well-understood data practices and don't inject conversational AI output — complete with chatbot response labels — directly into official documents. The presence of "Claude responded:" in a congressional document suggests a workflow where raw chatbot output was copied and pasted without adequate review, which is a different kind of concern entirely.
What Should Congressional Offices Do?
This incident offers a useful lesson for any government office navigating the integration of AI tools into daily workflows. Responsible AI use in legislative settings should include several key practices:
- Clear internal policies: Offices should establish written guidelines specifying which tasks AI tools may assist with and which are off-limits.
- Human review at every stage: Any AI-generated or AI-assisted content must be carefully reviewed and edited by a qualified human before it appears in any official document.
- Version control and documentation: Keeping records of how documents were produced — including any AI involvement — helps maintain accountability and supports transparency.
- Training for staff: Staffers using AI tools need to understand not just how to use them, but also their limitations and the risks of copying raw AI output without review.
Conclusion: A Warning Sign for the Future of AI in Government
Whether or not Rep. Anna Paulina Luna's staff actually used Claude to draft the substance of the defense amendment, the incident has already done something important: it has forced a public conversation about a practice that is almost certainly more widespread than most people realize. AI is already inside government buildings. The only question is whether the people using it are doing so carefully, transparently, and with appropriate safeguards in place.
As artificial intelligence continues to evolve and embed itself in professional workflows, the standards governing its use in democratic institutions need to keep pace. A chatbot's response text appearing in a congressional document — even accidentally — is a vivid reminder that the rules of the road for AI in government are still being written. The time to write them thoughtfully is now, before the next incident is something far more consequential than a spellcheck gone wrong.