Members of Congress are outsourcing their work to artificial intelligence without much, if any, oversight or limits.
Staffers and legislators are using AI chatbots to write speeches and news releases, manage mail from constituents, draft questions for congressional hearings, and even draft amendments, The Washington Post reports. The Senate and House have cleared staffers to use ChatGPT, Microsoft’s Copilot, and Google’s Gemini, with the House also approving Claude. Last year, the House bought 6,000 Copilot licenses, encouraging staffers to use them.
Senators and representatives regularly use the services themselves. Republican Senator Tommy Tuberville asks ChatGPT questions about issues such as Medicare and Medicaid while driving across his home state of Alabama.
“It’s kind of like you can sit there and talk to an expert,” Tuberville told the Post. “You don’t know whether or not they’re 100 percent correct. But they put you on the right path.”
Republican Representative Mike Flood asked a chatbot where private equity was buying single-family homes while working on a bipartisan housing bill last year. Democratic Senator Elizabeth Warren, who supports stricter regulations on tech companies, has used ChatGPT to look up demographic information.
Congressional staffers aren’t clear about if there are limits on using the bots. One staffer uses Copilot to draft bill amendments, while another fed all of his phone contacts and text messages going back years into Claude and told it to build a database of reporters, including what he pitched them, what they accepted, and what went badly. Claude also runs the communications budget of that congressional office, overseeing taxpayer-funded mailers and all ads sent to constituents.
That staffer isn’t sure what congressional rules allow for AI use.
“You’re still not supposed to put constituent data into AI,” he told the Post. “There are weird things around where the rules are. I don’t know all of them.”
In reality, Congress’s own rules are that AI can’t be used to make deepfakes, make personnel decisions, handle sensitive information such as constituent information, or finalize legislation. AI can be used to help draft a bill, but humans have the final review of it. And chatbots can be used to draft a response to a constituent, but aren’t allowed to be told who the constituent is.
Most of the staffers who spoke to the Post said they didn’t know these rules, and enforcing them is left to staffers self-policing or reporting their colleagues. This has already caused problems, such as when Republican Representative Anna Paulina Luna proposed an amendment to a defense bill earlier this summer.
One of her staffers included a summary of the proposed change with the amendment, and the first line began with “increasing example specific factors. Identical to H.R. 100 (118th Congress).11:25 AM????Claude responded:”, which was clearly lifted from a chatbot. Until rules and enforcement are tightened, incidents like this, and worse, will continue, and the public might not ever know.