Trump Administration’s Whole Government Ai Plans Leaked On Github

Updated We’re less than a month away from the Trump administration’s launch of an initiative to push AI across the entire federal government, based on a code repository eagle-eyed onlookers spotted on GitHub before it disappeared. 

The US General Services Administration (GSA, the federal government’s purchasing arm) and its Technology Transformation Services (TTS) group are working on an “ai.gov” website, according to a GitHub repository that vanished from the web shortly after we sent an email asking questions about it. (An archived backup is here.) The repository was previously reported by 404 Media.

From what we were able to gather before the feds presumably locked it down, AI.gov will serve as a hub for government agencies to begin adding AI to their operations, as was envisioned by TTS chief and Elon Musk ally Thomas Shedd when he took control of the team in late January. 

Shedd, whose professional career was largely spent as a software integration engineering manager at Tesla before being tapped to head the TTS, came to the government with AI top of mind. He reportedly wants GSA to operate like a software startup, and proposed a whole-of-government, AI-first strategy to automate much of the work done by federal employees today. 

Based on a staging link of the AI.gov site hosted on GitHub that has also been taken down (we have an archive copy for you, thankfully), Shedd’s mission will kick off in earnest on July 4 – the apparent launch date for the site, according to an issues thread from the now-hidden GitHub page.

AI.gov’s planned launch date is July 4, according to its developers – click to enlarge

Per the bare-bones staging implementation of ai.gov, the project has three components: A chatbot that’ll do … something; an “all-in-one API” that will allow agencies to connect their systems to models from OpenAI, Google, and Anthropic; and something called “CONSOLE,” which the page describes as a “groundbreaking tool to analyze agency-wide implementation.” 

From what we could gather based on the staging site, which didn’t include working copies of sub-pages, CONSOLE will allow agencies to monitor AI usage at their agencies in real time to see how employees are using tools and which ones they prefer. 

The staging site indicates that GSA is working with FedRAMP-certified vendors. According to API documentation from the GitHub page, AI.gov will serve AI models via Amazon Bedrock, most of which listed in the API documentation are known to be FedRAMP certified for government usage. But we did note the presence of a model from enterprise AI firm Cohere in the API documentation, and it doesn’t appear that Cohere has been FedRAMP certified. 

A list of some of the AI models AI.gov will push, according to API documentation – click to enlarge

GitHub documentation also indicates the site will publish model rankings, though we weren’t able to ascertain what the criteria would be.

The federal government has made a lot of noise about using AI lately, with the Trump administration and DOGE pushing for its adoption and trying to eliminate state-level regulations, while government agencies increasingly adopt it to replace employees eliminated in widespread layoffs and rely on it for critical decision making.

Experts have expressed concern, stressing that widespread adoption could create considerable security risks as AI systems gobble up confidential data and personally identifiable information about citizens. 

We contacted staffers involved in the AI.gov implementation, and Shedd, for comment but didn’t hear back from anyone, other than to see the repository door slammed shut. ®

Updated to add at 1502 UTC, June 11

A reader has been in touch to inform us that, while the ai.gov GitHub repository may appear gone, it wasn’t completely hidden – the GSA team just tossed it into a heap of archived projects. Those that want to take a poke around can still find it here.

Original Source