The AI Ghost in the Congressional Machine
Last week, I was scrolling through X (the platform formerly known as Twitter, because apparently we're doing that now) when I saw something that made me stop mid-scroll. A screenshot of a congressional amendment summary, complete with a telltale sign: the phrase "I am an AI assistant created by Anthropic." Yes, really.
According to www.theverge.com, Rep. Anna Paulina Luna (R-FL) found herself at the center of a very 2026 controversy. Her staff had apparently used Anthropic's Claude to help draft a summary for an amendment to a major defense funding bill. The response from Luna's office? A classic Washington two-step: "Yes we used AI, but only for spellcheck. No legislation is ever drafted with AI."
I've been covering tech and policy for 15 years. I've seen politicians stumble through congressional hearings on "the TikTok thing." I've watched senators ask Mark Zuckerberg if Facebook charges users. But this? This is a new flavor of awkward. It's the moment when the "we're just using it for spellcheck" defense meets the reality of what large language models actually are.
The Spellcheck Defense: A Closer Look
Here's the thing about the "spellcheck" argument. It's technically true that AI can check spelling. But Claude, ChatGPT, and their ilk don't just check spelling. They generate text. They rewrite. They suggest. They hallucinate. When you ask Claude to "help draft a summary," you're not running a grammar check. You're outsourcing the intellectual labor of summarizing a complex piece of legislation to a statistical model trained on the internet.
Let's be honest: if your staff is using AI to summarize a defense funding amendment, what else are they using it for? The line between "helping with phrasing" and "drafting the argument" is about as clear as the line between a lobbyist "suggesting language" and writing the damn bill. And we all know how that works.
According to www.theverge.com, the controversy started when accounts on X began sharing screenshots of the amendment summary that included the AI's own identification text. This is the kind of thing that happens when someone copy-pastes from an AI chatbot without checking. It's the digital equivalent of leaving the "Track Changes" on. Or, more aptly, leaving the receipt from the ghostwriter in the manuscript.
The Real Issue: Transparency vs. Efficiency
I've spent a lot of time on Capitol Hill. I've watched staffers run on four hours of sleep, chugging Diet Coke while writing talking points at 2 AM. I get the appeal of AI. I really do. If I could have a tool that helps me write a summary of a 1,200-page defense bill in 30 seconds instead of 3 hours, I'd be tempted too.
But here's where it gets uncomfortable. When a member of Congress says "NO Legislation is ever drafted with AI," they're drawing a bright line. Problem is, that line is already blurry. What about research? What about constituent correspondence? What about those "Dear Colleague" letters that everyone signs? What about the policy briefs that inform the actual drafting?
The issue isn't whether AI wrote the amendment text. The issue is that we have no framework for understanding when and how AI is being used in the legislative process. Zero. Zilch. Nada. We have more rules about what kind of pencils you can use to mark up bills than we do about AI assistance.
The Irony of It All
There's a delicious irony here that I can't help but point out. Congress is currently in a panic about AI. They're holding hearings about AI safety. They're drafting bills to regulate AI. They're worried about deepfakes and algorithmic bias and the existential threat of superintelligence. And yet, here they are, using the very same tools to write the laws that will regulate those tools.
It's like asking the fox to build the henhouse, but the fox is also using a drone to survey the property. Sure, the drone might help with the landscaping. But who's watching the fox?
I talked to a former congressional staffer (who asked not to be named because they still work in politics) who told me off the record: "Everyone uses it. Everyone. The only difference is whether they're smart enough to not leave the receipts." That's the dirty secret of AI in government right now. It's not a question of if it's being used. It's a question of who's getting caught.
What This Means for Accountability
Let me paint a scenario for you. Imagine a bill that includes a provision about AI training data. The provision was written by a staffer who used AI to "help with phrasing." The AI, trained on a dataset that reflects certain biases, subtly shapes the language in a way that benefits a particular industry. The staffer doesn't notice. The bill passes. The law has a hidden fingerprint that no one intended.
This isn't science fiction. This is the logical endpoint of the "spellcheck" defense. AI doesn't just correct your typos. It changes your vocabulary. It changes your framing. It changes what you think is normal. And when that happens in the context of legislation, you're not just getting a better word choice. You're getting a foreign perspective injected into the law.
Rep. Luna's office is probably right that no one used AI to write the actual amendment text. But the summary? The summary is what everyone reads. The summary is what gets shared with the press. The summary is what the CBO scores. The summary is, in many ways, more important than the text itself, because nobody reads the text.
The Broader Context: AI in Government
This incident is part of a larger pattern. We've seen AI used to write executive orders. We've seen AI used to draft responses to constituent letters. We've seen AI used to analyze public comments on proposed regulations. Every time, the response is the same: "We're just using it for efficiency." Every time, the question of accountability gets kicked down the road.
I'm not anti-AI. I use AI tools myself. I've written about AI for years. I think AI can be a powerful tool for good, including in government. But the lack of transparency is corrosive. If we don't know when AI is being used, we can't evaluate whether it's being used well. We can't audit for bias. We can't hold anyone accountable when things go wrong.
The irony is that AI could actually help with transparency. Imagine if every AI-assisted document came with a mandatory watermark. Imagine if there was a public log of AI usage in government. Imagine if we treated AI like we treat campaign contributions — something that must be disclosed.
What Should Happen Next
I have a suggestion. It's not radical. It's not even particularly innovative. But it would solve a lot of problems.
Every congressional office should be required to maintain a public log of AI usage. Not a detailed log of every prompt. Just a simple list: "We used Claude to draft a summary of H.R. 1234. We used ChatGPT to analyze constituent emails. We used Grammarly to check spelling." That's it. That's all it would take to build trust.
Second, the House and Senate should establish clear guidelines for what constitutes "drafting" vs. "assistance." If an AI suggests language and you accept it, that's drafting. If an AI corrects a typo, that's assistance. The line may be blurry in practice, but we can draw it somewhere and enforce it.
Finally, we need a conversation about whether we want our laws to be written by humans or by machines. I think the answer is obvious: humans. But the technology is moving faster than our ability to have that conversation. By the time we figure out the rules, the AI will have already written them.
A Personal Observation
I'll leave you with this. Last week, I tried to write a letter to my congressperson about AI policy. I started drafting it myself. Then I thought, "Why not use AI to make it more persuasive?" So I fed my draft into Claude and asked it to "improve the language." It came back with a version that was more polished, more convincing, and... completely hollow. It had the rhythm of persuasion without the substance. It sounded like a politician.
I deleted the AI version. I wrote my own letter. It was less elegant. It had typos. But it was mine. And if we lose that — the ability to speak in our own voice, even in government — we've lost something fundamental.
What happens when the people writing our laws stop sounding like people? What happens when the only authentic voice in government is the one being flagged for spellcheck?
Originally reported by www.theverge.com. Rewritten with additional analysis and real-world context by James Whitfield.
