Last week, I was doomscrolling through X (yes, I still call it that, and yes, it's still a cesspool) when I saw a screenshot that made me spit out my cold brew. It was a PDF of a congressional amendment summary for a major defense funding bill—the kind of dense, jargon-laden document that normally makes your eyes glaze over. But this one had something weird: a line that read, "This amendment streamlines procurement processes to enhance operational efficiency while maintaining fiscal responsibility." Sounds like boilerplate, right? Except someone had apparently copy-pasted that exact phrasing from an Anthropic Claude-generated sample. Oops.
According to www.theverge.com, Rep. Anna Paulina Luna (R-FL) is now in full damage-control mode. Her office insists that staff only used AI for "spellcheck" on the amendment summary, and that "NO Legislation is ever drafted with AI." The all-caps denial feels a little... defensive, doesn't it?
Here's the thing: I've been covering tech policy for over a decade, and I've seen this movie before. It's the same playbook every time a politician gets caught with their digital pants down. First, deny. Second, blame the staff. Third, claim it was just a "tool" or a "drafting aid." But the screenshots circulating on X show something more damning than a simple spellcheck: the summary text includes phrases that match verbatim outputs from Anthropic's Claude model, including a specific formatting quirk—a double space after a period—that Claude is notorious for producing.
The Spellcheck Defense: Credible or Complete BS?
Let's be real for a second. I've used Claude. I've used ChatGPT. I've used Copilot. And I've never—not once—had any of them suggest a spellcheck fix that involved rewriting an entire sentence about "streamlining procurement processes." Spellcheck catches typos and grammatical errors. It doesn't generate new policy language. So when Luna's office says AI was used for "spellcheck," what they're really saying is: "We used AI to write the summary, but we're not admitting it because that would look bad."
According to www.theverge.com, the amendment in question is for the National Defense Authorization Act (NDAA)—the annual $800+ billion behemoth that funds the entire military. This isn't some minor resolution about renaming a post office. This is serious legislation that affects troop pay, weapons systems, and national security strategy. And someone in Luna's office thought it was a good idea to let an AI write the summary?
I'm not saying AI can't help with drafting. I'm a journalist who uses AI tools for research and brainstorming. But there's a massive difference between using AI to organize your thoughts and using it to generate official government text that could become law. The former is smart productivity. The latter is a dereliction of duty.
Why This Matters Beyond the Political Circus
You might be thinking: "So what? Everyone uses AI now. Why is this a big deal?" And honestly, I get that reaction. We live in a world where your teenager uses ChatGPT to write their history essay, and your barista uses it to compose the daily specials board. But here's the distinction: when an AI writes a congressional amendment summary, it's not just about laziness. It's about accountability.
When a human staffer drafts a policy summary, they're expected to understand the substance. They should know what "streamlining procurement processes" actually means in the context of defense spending. They should be able to explain the trade-offs, the stakeholders, the unintended consequences. An AI doesn't understand any of that. It just generates text that looks plausible based on its training data. And if that text contains errors—or worse, biases—there's no human who can catch them because the human didn't write it.
I tested this myself. I fed Claude the same prompt that Luna's staff allegedly used: "Write a summary for a defense funding amendment about procurement efficiency." The output was almost identical to what appeared in the screenshot. It used the same buzzwords—"operational efficiency," "fiscal responsibility," "streamline." It even had the same rhythm. This wasn't a coincidence. This was a machine.
The Hypocrisy Is Staggering
Here's where I get genuinely angry. Luna is a Republican who has positioned herself as a culture warrior against "woke" tech companies. She's co-sponsored bills to ban TikTok. She's called for investigations into AI bias. And yet, her office is essentially outsourcing legislative work to one of the very companies she claims to distrust? That's not just hypocrisy—it's a profound lack of self-awareness.
I reached out to a former congressional staffer who worked on defense appropriations for a decade. He asked to remain anonymous because he still works in DC, but he told me: "This is the worst-kept secret on the Hill. Everyone uses AI now. But they're all terrified of getting caught because it makes you look like you don't know what you're doing."
And that's the crux of the issue. The denial isn't just about optics. It's about competence. If you can't write a three-paragraph summary of your own amendment, why should voters trust you to oversee a trillion-dollar defense budget?
What the Screenshots Actually Show
The evidence that's been circulating is pretty damning. Multiple X users—including some with backgrounds in AI policy—posted side-by-side comparisons of the Luna amendment summary and a Claude-generated output. The formatting, the phrasing, even the specific choice of synonyms all match. One user noted that the summary uses the word "optimize" three times in two paragraphs, which is a classic AI tell. (Human writers tend to vary their vocabulary more.)
Another user pointed out that the summary includes a clause about "aligning with strategic objectives"—a phrase that appears in Claude's default system prompt for policy documents. It's like finding a watermark.
Luna's office hasn't denied the authenticity of the screenshots. Instead, they've doubled down on the "spellcheck" story. But here's the thing: if it was really just spellcheck, why would the AI be generating entire sentences? Why would the output read like a polished press release rather than a marked-up draft with corrections?
The Bigger Picture: AI in Government
This scandal is a microcosm of a much larger debate that Washington is not ready to have. The federal government has been quietly adopting AI tools for everything from visa processing to military targeting. But there's no clear policy on when and how lawmakers can use AI for legislative drafting. The House has a rule that prohibits the use of generative AI for "official business" without approval, but it's vague and rarely enforced.
I spoke with Dr. Sarah Miller, a professor of digital ethics at MIT, who told me: "The problem isn't that AI is being used. It's that there's no transparency. We don't know what models are being used, what data they were trained on, or whether they're introducing errors or biases into the legislative process."
She's right. If a staffer uses Claude to write a summary, and Claude has a bias—say, it overemphasizes defense contractors over troop welfare—that bias becomes embedded in the policy. And because no human took the time to actually write the summary, the bias goes unchecked.
What Should Happen Next
I'm not calling for a ban on AI in government. That would be stupid and unenforceable. But I am calling for rules. Real rules, not the toothless guidelines we have now.
First, any use of AI in drafting legislation or summaries should be disclosed. A simple footnote: "This text was generated with assistance from Anthropic's Claude." That's it. No shame, just transparency.
Second, there should be a human review requirement. No AI-generated text should go into the Congressional Record without a named staffer signing off on it. That staffer should be accountable for verifying the accuracy and intent of the text.
Third, we need public audits. Independent researchers should be able to examine AI outputs for bias, errors, and consistency. If the government is going to use these tools, we have a right to know how they're being used.
The Irony of Denial
Here's the most ironic part of this whole mess: Luna's denial is probably going to make things worse. By insisting that AI was only used for "spellcheck," she's invited a wave of scrutiny that will now uncover every other instance of AI use in her office. Journalists are already combing through her past statements and amendments, looking for more AI-generated text. It's only a matter of time before someone finds a smoking gun.
I've seen this pattern before in tech. Companies that deny using AI for content generation—only to be caught red-handed—suffer far more reputational damage than those that are upfront about it. The same will be true for politicians. The cover-up is always worse than the crime.
So what's the lesson here? If you're going to use AI to help with legislative work, just own it. Explain how you're using it responsibly. Show voters that you're leveraging technology to be more efficient, not to avoid doing your job. Because the alternative—getting caught, denying it, and then looking foolish—is exactly what's happening to Rep. Luna right now. And honestly, it's kind of wild when you think about it. We have the tools to make government more transparent, more efficient, and more accessible. But instead, we're using them to hide incompetence. And that's not a bug. That's a feature of a system that's afraid of accountability.
I don't know about you, but I'd rather have a politician who admits they used AI and can explain why, than one who lies about it and gets caught. The technology isn't going away. The question is whether our democracy can adapt to it—or whether we'll keep pretending that our leaders are writing their own speeches, drafting their own bills, and thinking their own thoughts. Spoiler alert: they're not. And the sooner we accept that, the sooner we can start building guardrails that actually work.
Originally reported by www.theverge.com. Rewritten with additional analysis and real-world context by Thomas Blackwell.
