Key Takeaways
- As of September 23, 2026, US Senate staff are authorised to use only three AI chat tools for official work: OpenAI’s ChatGPT Enterprise, Google Gemini Chat and Microsoft Copilot Chat, approved by the Senate Sergeant at Arms in March 2026.
- Agentic AI tools, including Anthropic’s Claude Code, are explicitly barred because Senate policy prohibits platforms that can independently access internal drives, email, Teams chats or other Senate resources.
- On the same day NPR reported these restrictions, Senators Bernie Sanders and Greg Casar introduced the Ban Artificial Superintelligence Act, which would pause advanced AI development pending the creation of a new cabinet-level Department of Artificial Intelligence. Senate staff writing AI legislation cannot use the AI they are legislating. While OpenAI’s ChatGPT Enterprise, Google Gemini Chat and Microsoft Copilot Chat gained approval for Senate use in March 2026, more capable agentic systems remain off-limits entirely, and the gap between what staff can access and what Congress is debating has become hard to ignore.
Three Tools, One Licence Each
The March 2026 authorisation came from the Senate Sergeant at Arms’ Chief Information Officer and covers three platforms: ChatGPT Enterprise, Google Workspace with Gemini Chat and Microsoft Copilot Chat. Staff can use them for drafting and editing documents, summarising information, preparing talking points and conducting research. Each Senate employee receives one generative AI licence at no cost, for either Gemini Chat or ChatGPT Enterprise. Microsoft Copilot is notable for its integration into the Microsoft 365 environment already in use across Senate offices.
The House of Representatives has gone further, also approving Anthropic’s Claude Pro for official use alongside the same three chatbots.
Where Agentic Tools Are Blocked
Senate policy, as reported by NPR on September 23, 2026, explicitly bars staff from agentic AI systems, tools that can autonomously complete multi-step tasks rather than simply respond to queries. The prohibition is specific: platforms “cannot independently access internal Senate drives, shared folders, email, Teams chats, or other Senate resources,” according to Senate rules. Tools with those capabilities, including Anthropic’s Claude Code, are either still under evaluation or not authorised. The security concern is straightforward: autonomous access to sensitive legislative systems carries risks that basic chat tools do not.
Regulating What They Cannot Use
Adam Kovacevich, founder and CEO of the tech policy group Chamber of Progress, said lawmakers are often writing rules for technology they have never used, according to NPR.
On that same day, Senators Bernie Sanders and Greg Casar introduced the Ban Artificial Superintelligence Act, which would halt advanced AI development until a new cabinet-level Department of Artificial Intelligence could establish safety rules. The bill reflects genuine alarm about autonomous AI systems. It also illustrates the disconnect plainly: the senators drafting rules for the most advanced AI in existence are, institutionally, restricted to chat interfaces that cannot open a shared folder.
Originally published at https://autonainews.com/senate-staff-get-three-approved-ai-tools-but-agentic-systems-stay-banned/
Top comments (0)