The Senate has approved the use of AI chatbots—ChatGPT, Gemini, and Copilot—for official staff work to assist with research, drafting, editing, and briefing materials, under a governance-focused framework.

The approval comes with security safeguards: Copilot Chat does not access Senate data unless included in prompts and operates within the Microsoft 365 Government environment.

Claude (Anthropic) and Grok (xAI) are not on the approved list, with Claude under internal review amid policy disputes over mass surveillance and autonomous weapons.

Details on timing, administrators, or scope are not provided beyond the central approval.

A separate development: Oracle and OpenAI reportedly dropped plans to expand a flagship AI data center in Texas as part of Stargate, amid financing negotiations.

The broader Washington debate over AI security and surveillance concerns continues, including tensions with Anthropic and the Pentagon.

Ongoing questions remain about adoption breadth and internal policy limits, as offices and committees may set their own rules.

Critics warn of misinformation, bias, and privacy risks, underscoring the need for ongoing human oversight of AI outputs in government.

The context shows increasing government and military use of AI, with industry players navigating regulatory and policy tensions, including Anthropic’s supply-chain risk labeling.

Microsoft said it was reviewing the approval, while Google and OpenAI declined to comment; Reuters notes Microsoft awaits final approval and other companies have not confirmed rollout.

Official guidance advises against handling sensitive or classified information in AI tools; staff should avoid PII or security details, with multi-layer approvals for complex tasks.

Copilot Chat will not automatically search internal drives, emails, Teams chats, or shared folders, as the memo states.