House AI rules left unenforced, individual offices left to police themselves

The US House of Representatives bought 6,000 Microsoft Copilot licenses in 2025. It also approved Claude for official use. A Washington Post investigation, drawing on interviews with more than two dozen lawmakers and congressional staff, paints a picture of Capitol Hill increasingly dependent on AI chatbots for core legislative functions: drafting speeches, responding to constituents, summarizing complex bills, sorting through mountains of correspondence. When the rough draft becomes the final draft In June 2026, Rep. Anna Paulina Luna’s office filed an amendment that turned out to contain raw AI-generated output, complete with erroneous content that apparently nobody caught before submission. Luna defended her office’s use of Claude after the incident. Staffers report meaningful efficiency gains on research, correspondence sorting, and boilerplate text. But the outputs frequently require heavy revision, which means the time savings are partially eaten up by quality control. Or worse, the quality control step gets skipped entirely. The oversight gap Both the House and Senate have issued guidance on how staff should use AI tools, including prohibitions on sharing sensitive constitue...