Congress Widely Adopts AI Chatbots With 6,000 Copilot Licenses as Oversight Largely Fails
Updated
Updated · The Washington Post · Aug 13
Congress Widely Adopts AI Chatbots With 6,000 Copilot Licenses as Oversight Largely Fails
3 articles · Updated · The Washington Post · Aug 13
Summary
A garbled NDAA amendment summary pasted by Rep. Anna Paulina Luna’s office exposed how routinely congressional staff use chatbots to draft legislative materials, with Luna saying “most staff use it.”
Across Capitol Hill, members and aides now use ChatGPT, Claude, Grok, Gemini and Copilot for speeches, constituent mail, hearing prep and amendments, after the House bought 6,000 Copilot licenses and put roughly half into use.
House and Senate rules bar uploading sensitive constituent data, deepfakes, personnel decisions and letting AI finalize legislation, but interviews and internal policies show those limits are poorly understood and rarely enforced.
No one interviewed could cite a formal punishment for AI-rule violations, leaving individual offices to set their own guardrails while some aides push further by building reporter databases or trying to mimic a lawmaker’s voice.
That rapid adoption leaves Congress regulating the same tech companies whose tools now shape its daily work, raising broader questions about how much legislative judgment lawmakers are willing to outsource.