US Senate authorises ChatGPT Gemini and Microsoft Copilot for official work
Sergeant at Arms memo formalises AI drafting and research across offices, Claude left under evaluation as vendors become workflow infrastructure
Images
On Monday the US Senate’s Sergeant at Arms told offices they may use three generative AI tools—OpenAI’s ChatGPT Enterprise, Google Workspace with Gemini Chat, and Microsoft Copilot Chat—for official work, according to a memo obtained by Business Insider and first reported by The New York Times. The note explicitly lists tasks such as drafting and editing documents, summarising information, preparing talking points and briefing material, and conducting research and analysis. Copilot is positioned as the default option: it is already integrated into the Microsoft 365 environment used across the Senate and is available to staff “at no cost,” while ChatGPT Enterprise and Gemini require Senate licences.
The change formalises what many offices had already been doing informally. Several senators told Business Insider last year that staff were experimenting with AI for research and drafting, with office-by-office rules still evolving. The memo is an institutional admission that the marginal cost of producing text—briefs, memos, constituent responses, and internal summaries—has fallen, and that Congress intends to capture that productivity gain.
But the memo also sketches the new dependency chain. If staff use these systems to generate first drafts and compress complex material into bullet points, the practical standard for “how Senate work is written” shifts toward whatever the dominant vendors optimise for. Microsoft benefits from being embedded in the workflow, Google and OpenAI from being authorised alternatives, and the Senate from being able to push more output through the same headcount. The cost is harder to price: prompts become a new layer of sensitive material, and “drafting and summarising” is exactly where legislative intent, negotiation positions, and politically awkward caveats tend to appear before they are scrubbed.
The Senate’s guidance tries to contain that risk by emphasising boundaries. It says Copilot Chat does not access Senate data unless a user explicitly shares it in a prompt and does not search internal drives, shared folders, email, or Teams chats on its own. It also notes Copilot runs in Microsoft’s secure government cloud and meets federal and Senate cybersecurity requirements. That framing shifts responsibility from the platform to the employee: the tool is safe, the memo implies, as long as staff do not paste the wrong thing.
What the memo does not explain is why Anthropic’s Claude—already approved for House use, according to the POPVOX Foundation—was left out. An internal Senate IT message viewed by Business Insider says Claude is among tools still under evaluation. The omission lands amid a broader political dispute: the Trump administration has ordered federal agencies to stop using Anthropic’s technology after the company restricted certain uses, including mass surveillance and autonomous weapons, though that order does not apply to Congress.
For now, the Senate is standardising on three vendors for the daily work of turning information into language. The memo promises each Senate employee one AI licence for either ChatGPT Enterprise or Gemini at no cost, with details to follow within 30 days.