Congress Has a New Invisible Lobbyist: The Rise of the Political Chatbot

Congress now has a lobbyist that never registers, never discloses a client, and never leaves a paper trail, and it sits open in a browser tab on nearly every staffer's desk. The House Office of Legislative Counsel, the institution responsible for turning policy ideas into legally binding text, has reportedly been inundated with AI-generated bill proposals so riddled with incorrect statutory references and sloppy drafting that its lawyers now spend more time fixing them than they would spend writing the bills from scratch. Around the same time, it became clear how far this had spread. Members of Congress and their staff are increasingly using systems such as ChatGPT, Claude, Copilot, and Gemini for policy research, legislative drafting, speeches, and constituent replies.

On the surface, this looks like any workplace chasing efficiency. But Congress is not any workplace. The information lawmakers receive, the way problems get framed, and the exact language that ends up in statute can affect millions of people. So the real question is not whether chatbots make congressional offices faster. It is what happens once the systems helping policymakers research, frame, and write legislation become an active participant in the political process itself.

The appeal of AI on Capitol Hill is easy to understand. Legislative work is slow, technical, and dependent on a deep familiarity with existing law, and staff are expected to research issues, draft bills and amendments, prepare hearing questions, and answer constituents, all while navigating a legal system where small wording differences change how a law gets interpreted. Both chambers have cleared members and staff to use certain chatbots for official work, and the House alone purchased 6,000 Microsoft Copilot licenses, with roughly half of them currently in use. These tools can summarize long documents, organize research, and generate drafts in seconds, which makes them attractive to offices already stretched thin on time and staffing.

The trouble starts once AI moves past routine administrative help. Asking a chatbot to summarize a document is one thing. Asking it to explain a policy problem, weigh regulatory options, or draft legislative language is another, because the system stops retrieving information and starts shaping it, selecting what gets included, what gets left out, and how the whole issue gets framed before a human even weighs in.

That distinction between retrieving information and mediating it is the whole story here. A search engine hands users a pile of sources to sort through themselves. A chatbot hands back one clean answer, doing part of the thinking a staffer would otherwise have done. The staffer can still edit or reject that output, but the frame they start from has often already been set by the machine.

Lawmakers have always operated in a crowded room of people trying to shape how they see an issue, lobbyists, advocacy groups, think tanks, expert witnesses. What separates a chatbot from all of them is visibility. A lobbyist represents a named client. A chatbot represents no one in particular, yet it can still steer what a staffer treats as the obvious risks, the realistic options, or the models worth copying from other countries, all before the political debate even opens.

None of this means AI companies are steering Congress toward outcomes that benefit them, and there is no evidence of anyone rigging outputs for that purpose. But influence does not require a conspiracy. Framing alone can decide which solutions look reasonable or urgent, especially when policymakers increasingly lean on AI to process material they don’t have the time or specialized knowledge to check themselves. In that sense, the political chatbot behaves like an invisible lobbyist, not one pushing an agenda openly, but one shaping the starting conditions of political thought from the inside. The real danger isn’t that AI will start telling Congress what to do. It’s that lawmakers may simply get used to beginning every research and drafting task from whatever frame the AI hands them first, until the chatbot stops being a tool that assists political thinking and becomes the channel political thinking runs through.

The strain on the Office of Legislative Counsel shows exactly why fast AI adoption can create as many problems as it solves. The office is one of Congress’s key quality-control checks, making sure bill language actually matches what a lawmaker intended and fits inside existing law. Its workload was already heavy before generative AI arrived. Under the previous Congress, 61 attorneys and 19 support staff prepared more than 30,000 bills, nearly three times the number that were ever introduced. In just the first 60 days of the current Congress, the office fielded 5,623 legislative requests, a 72% jump from two years earlier.

AI widens that gap because it collapses the time needed to generate a bill draft to almost nothing, letting an office, advocacy group, or outside organization produce several versions of legislation in minutes, even without the specialized legal training drafting has always required. But while AI can produce text instantly, checking whether that text is accurate, internally consistent, and legally sound is still a slow, human process. Call it an AI verification gap. The ability to generate proposals is scaling up fast, while the institutional capacity to check them barely moves, and the space between those two speeds keeps growing.

That gap is already visible in what lands on lawyers’ desks. AI-drafted bills arrive with wrong statutory citations, mismatched terminology, and vague provisions, and fixing them can take longer than writing the bill would have taken in the first place. Rep. Joe Morelle put the stakes plainly, warning that “AI cannot replace the skilled and dedicated professionals in the Office of Legislative Counsel.” Rep. Norma Torres was just as direct, cautioning that “relying too heavily on AI-generated bill text could introduce errors.”

The mistakes matter because legislative drafting lives or dies on small distinctions. Whether aid is structured as a tax credit, a deduction, or a grant changes who actually benefits and how. A sloppy definition of “state” could accidentally exclude the District of Columbia or tribal nations from a federal program. A wrong citation to the US Code can leave agencies and courts guessing at what Congress actually meant.

The deeper risk goes beyond bad bills, though. Wide AI use could slowly reshape how staff build expertise in the first place. Drafting isn’t just converting an idea into legal language, it forces staff to work through definitions, exceptions, legal authority, and unintended consequences, and that struggle is how institutional knowledge gets built. A staffer who drafts a proposal from scratch has to reason through all of it. A staffer who feeds a goal into a chatbot gets back a structure someone, or something, else already built, with the hard choices about framing and priority already made.

Reviewing and editing that output is still possible, but leaning on AI-generated first drafts again and again can flip the order of operations, from building a policy framework and then writing it into law, to starting from whatever framework the AI hands over and editing from there. That is institutional deskilling. If AI keeps absorbing more of the research, synthesis, and drafting that used to build staff expertise, Congress could end up more dependent on the tool while less able to catch it when it’s wrong, a feedback loop where doing less of the underlying work by hand makes people worse at spotting the technology’s mistakes. The challenge, then, isn’t just stopping AI hallucinations from reaching the floor. It’s making sure the technology doesn’t erode the human expertise needed to catch those hallucinations in the first place.

There’s a second complication, which is who actually owns these systems. Congress isn’t relying on some purpose-built public tool for legislative work. It’s relying on products from Microsoft, OpenAI, Google, and Anthropic, companies whose own activities Congress may eventually have to regulate. That doesn’t mean these companies control the legislative process or are tilting outputs in their own favor. The concern is structural, not conspiratorial.

Unlike ordinary office software, AI increasingly performs tasks that used to require a human mind, researching, comparing arguments, spotting patterns, proposing solutions. That makes it a form of privately owned cognitive infrastructure, meaning institutions are starting to depend on privately built systems for the basic work of processing information and forming ideas. That dependency creates its own risk. Congress can remain formally independent of these companies while still leaning on their products for daily work, and the more staff build habits and workflows around a particular system, the harder it becomes to walk away from later.

That tension is sharpest given Congress’s actual job here, since it’s also responsible for writing the rules that will eventually govern this industry. A staffer researching AI regulation might use a privately built chatbot to understand the risks tied to the very industry that built it. Again, that’s not proof of bias. But it raises a real question about how independent an institution can stay once private systems are embedded in how it gathers and processes knowledge in the first place.

Traditional political influence comes with at least some transparency built in. Lobbyists register. Corporate spending gets tracked. Experts can be asked about conflicts of interest. AI offers none of that by default. A staffer can get a confident, polished answer with no visibility into why certain sources or arguments were emphasized and others left out. That doesn’t make the answer wrong, but it creates an accountability problem that’s much harder to trace than a traditional lobbyist’s interest.

This pattern reaches well past Congress. AI is becoming the first point of contact between people and information across education, journalism, and everyday life, and instead of sorting through competing sources themselves, users increasingly just ask a chatbot to do the sorting. That’s genuinely convenient, but it also concentrates real power in whichever systems decide how information gets organized and presented. If that continues, a small number of AI platforms could become de facto gatekeepers of political and social knowledge, not by pushing one ideology on everyone, but simply because millions of people, and the institutions meant to govern them, increasingly start from the same handful of privately built systems.

Social media reshaped how politicians talk to voters. Recommendation algorithms reshaped what information people encounter. AI goes a step further, since it doesn’t just filter information, it helps produce and synthesize it. That’s why this conversation shouldn’t get reduced to whether a chatbot can draft a bill faster than a human staffer can. The real issue is what happens once AI gets embedded at the earliest point of political decision-making, shaping how a problem gets understood before anyone even starts debating solutions.

The immediate fallout is already visible, error-ridden bills, an overwhelmed legal office, and a workload that keeps climbing. The longer-term stakes are bigger, because AI is shifting from something Congress merely regulates into something that helps determine how Congress itself researches, communicates, and legislates.

The real question isn’t whether AI will replace lawmakers or start writing laws on its own, elected officials and legislative counsel still hold formal responsibility for what becomes law. It’s whether Congress can hold onto enough independent expertise to keep AI as a tool it uses, rather than an invisible structure political decisions increasingly pass through. In that sense, the chatbot earns the “invisible lobbyist” label not because it pushes an agenda for a paying client, but because it can shape the environment political choices get made in without any of the disclosure rules that govern every other form of political influence. Its biggest impact may not land on the day of a vote, but far earlier, the moment a staffer asks it what matters, what the options are, and how to think about the problem at all.

As Congress keeps building privately made AI into its own work while trying to write the rules for that same industry, protecting independent human judgment stops being a matter of personal caution and becomes an institutional problem in its own right. The real risk was never lawmakers suddenly handing power to a machine. It’s a slower slide, where AI-generated answers become the default starting point for research and drafting, until the line between independent judgment and algorithmically shaped judgment gets hard to see at all.

Brown, Hayes. 2026. ‘Congress Is Getting Bogged Down by AI Slop’. MS NOW, August 19. https://www.ms.now/opinion/congress-house-senate-staff-ai-use.

Congress.Gov. 2026. ‘AI Chatbots as Companions: Overview, Uses, and Considerations for Congress’. Congress.Gov. https://www.congress.gov/crs-product/R49189.

Dahlkamp, Owen. 2026. ‘AI Slop Is Swamping a House Office That Drafts US Laws’. Politico, Politico, August 17. https://www.politico.com/news/2026/08/17/ai-slop-lawmakers-congress-01008376.

Liss-Roy, Anna. 2026. ‘Chatbots Are Doing the Work of Congress With Little Oversight’. The Washington Post, August 13. https://www.washingtonpost.com/politics/2026/08/13/chatbots-are-doing-work-congress-with-little-oversight/.

Rashid, Hafiz. 2026. ‘The Wild—and Unregulated—Ways Lawmakers Are Using AI Chatbots’. The New Republic, August 13. https://newrepublic.com/post/214291/congress-ai-chatbots-write-bills-tommy-tuberville.

The New York Times. 2006. ‘Opinion | Congress’s $2 Billion Typo’. Nytimes, The New York Times, February 23. https://www.nytimes.com/2006/02/23/opinion/congresss-2-billion-typo.html.