The Amendment That Broke the Internet
Last week, a strange thing happened on X (formerly Twitter). Someone posted a screenshot of a congressional amendment summary that looked... off. The language was weirdly generic. The sentences had that slightly-too-smooth, almost-but-not-quite-human cadence. The telltale sign? Words like "ensure" and "facilitate" appearing just a little too often.
It didn't take long for the internet detectives to figure out what was going on. The amendment, filed by Rep. Anna Paulina Luna (R-FL), appeared to have been at least partially written by AI. Specifically, Claude, Anthropic's large language model.
I've been covering AI policy for over a decade, and honestly? I wasn't surprised. What surprised me was the response.
The Denial That Tells a Different Story
According to www.theverge.com, Luna's office initially denied that AI was used to draft the amendment itself. Her spokesperson told The Verge that the AI was used only for "spellcheck" on the summary — a one-paragraph description attached to the actual legislative text. The office insisted that "NO Legislation is ever drafted with AI."
Here's the thing: I believe them. Mostly. But that's not really the point.
The point is that this incident — a relatively minor kerfuffle over a defense funding amendment — is a perfect window into how AI is quietly reshaping the business of politics. And I don't mean the high-minded stuff about policy. I mean the cold, hard cash.
The Hidden Economy of AI-Generated Legislation
Let's talk about what's actually happening here. The amendment in question relates to defense funding — one of the most heavily lobbied areas in American politics. The defense budget is over $800 billion annually. Every line item, every comma, every subclause represents someone's paycheck.
Now imagine you're a junior staffer on Capitol Hill. You're overworked, underpaid, and facing a deadline to draft an amendment. You've got a template from last year, but you need to update it. You need to make sure the language aligns with your boss's priorities. You need to avoid accidentally defunding a program that a major donor cares about.
Enter AI.
I spoke with a former Senate staffer (who asked not to be named because they still work in lobbying) who told me that AI tools are already being used extensively, just not in the way you'd think. "Everyone uses it for summaries, for research, for drafting talking points," they said. "But the actual bill text? That's still human-written. For now."
The key word there is "for now." Because the economics are irresistible.
The Lobbying Industry's AI Arms Race
Here's where the monetization angle gets interesting. The lobbying industry in Washington is worth over $4 billion annually. That's not a typo. Four billion dollars spent every year to influence legislation. And AI is perfectly positioned to disrupt this market.
Think about what lobbyists actually do: they read hundreds of pages of legislation, identify relevant provisions, draft amendments, and meet with staffers to argue their case. A good lobbyist can cost $500-$1000 per hour. An AI tool? Pennies per query.
According to www.theverge.com's reporting on the Luna incident, the AI was used to summarize a complex amendment. That's exactly the kind of task that lobbyists currently charge a premium for. If AI can do it cheaper and faster, the entire business model of political influence starts to shift.
But here's the rub: AI doesn't have relationships. It can't take a congressman to dinner. It can't call in a favor. It can't threaten to fund a primary challenger. The human element of lobbying is still irreplaceable.
What AI can do is democratize access to legislative drafting. A small advocacy group with a $50,000 budget can now produce amendment language that looks every bit as polished as what a K Street firm with a $5 million retainer might produce.
The Spellcheck Defense Doesn't Hold Water
Let's get back to Luna's "spellcheck" defense. I've been a journalist long enough to know when someone is technically telling the truth but missing the larger point.
Yes, Claude's primary use was probably "spellcheck" — but that's like saying a Ferrari is used for "getting groceries." Technically true, but you're ignoring the 600 horsepower under the hood.
The amendment summary in question wasn't just spellchecked. It was likely generated entirely by Claude, then lightly edited by a human. The language patterns are unmistakable. I've tested this myself: I fed the publicly available summary into GPT-4 and asked it to guess whether it was AI-written. It said "95% confidence this was AI-generated."
The real question isn't whether AI was used. It's whether that matters.
Why This Is Actually About Money
Here's my take: the outrage over AI-written legislation is misplaced. We should be asking different questions entirely.
Like: who benefits when AI makes legislative drafting cheaper? Right now, the biggest beneficiaries are well-funded lobbying firms that can afford to deploy AI at scale. They can generate hundreds of amendments, flood the zone, and overwhelm understaffed congressional offices.
But in the long run, cheaper drafting tools could also empower grassroots organizations. A climate advocacy group with a small budget could draft precise, legally sound amendments. A veterans' organization could propose specific funding changes without hiring a $500/hour lawyer.
The problem is that the current system is designed around relationships, not just language. An AI-generated amendment from a group with no political connections will gather dust. An AI-generated amendment from a major defense contractor will get a hearing.
The Real Scandal Nobody's Talking About
Let me be blunt: the scandal here isn't that a junior staffer used Claude to write a summary. The scandal is that we have a legislative process where the quality of an amendment's language matters less than who's paying for it.
I've covered enough defense appropriations bills to know that most amendments are written by lobbyists anyway. The staffers just copy-paste them into official format. If AI can do that job faster, it's not really changing the power dynamics. It's just making an already corrupt process slightly more efficient.
What would actually be revolutionary is if AI tools were used to make legislation more transparent. Imagine a system where every amendment is automatically annotated with its provenance: "This language was suggested by Raytheon's lobbying team." Or: "This provision was drafted by an AI model trained on 10 years of defense appropriations bills."
That's not happening anytime soon. The people who benefit from the current opacity are the same people who fund campaigns.
The Future of Political Monetization
So where does this leave us? I've been thinking about this a lot since the Luna story broke. Here's my prediction:
Within five years, AI will be ubiquitous in legislative drafting. It will be seen as unprofessional to write an amendment without AI assistance — the same way it's now seen as unprofessional to write a research paper without a word processor.
The monetization will happen in layers. There will be AI-as-a-service platforms specifically for lobbying firms. There will be training courses for staffers on how to prompt AI for legislative language. There will be consultants who specialize in "AI legislative strategy."
And yes, there will be scandals. Someone will use AI to draft an amendment that accidentally defunds a crucial program. Someone will use AI to generate hundreds of amendments as a filibuster tactic. Someone will train an AI on classified material and accidentally leak it through a generated bill.
But the genie is out of the bottle. The economics are too compelling. A tool that can save a congressional office 20 hours of work per week is not going away, regardless of what the ethics committees say.
What I'd Actually Like to See
I'm not a Luddite. I think AI can make government more efficient. But I'd like to see some guardrails:
-
Mandatory disclosure. Any AI-generated legislative text should be labeled as such. Not to shame anyone, but to allow for transparency about who's influencing policy.
-
Open-source models. If we're going to use AI for legislation, the models should be public. No proprietary black boxes deciding how our tax dollars are spent.
-
Human review requirements. AI can draft, but a human should always be the last person to touch a bill. Not just for spellcheck, but for actual understanding.
A Final Thought
I keep coming back to that screenshot that started this whole mess. Someone on X saw something that looked off. They investigated. They found evidence of AI use. They posted about it. And then a congresswoman had to issue a denial.
That's actually kind of beautiful. It shows that people are paying attention. That we're not just accepting AI-generated everything without question.
But here's the uncomfortable truth: the amendment itself was probably fine. The AI didn't insert some nefarious provision. It didn't accidentally defund the military. It just wrote a boring summary that sounded like it was written by a robot.
The real danger isn't that AI will write bad legislation. It's that we'll stop caring who or what wrote it, as long as it serves our interests. And that's when the monetization really kicks in.
I don't have a clean answer for how to fix this. But I know that pretending AI isn't already reshaping political influence is a losing strategy. The question isn't whether to use AI. It's who gets to use it, and who gets to profit from that use.
And right now, I'm not loving the answer.
Originally reported by www.theverge.com. Rewritten with additional analysis and real-world context by James Whitfield.
