The so-called "Department of Government Efficiency" (DOGE), led by Elon Musk, has rolled out a proprietary chatbot called GSAi to 1,500 federal employees at the General Services Administration (GSA). This move to automate tasks previously performed by humans comes amid DOGE's ongoing reduction of the public sector workforce.
GSAi is intended to support "general" tasks, similar to commercial tools like ChatGPT or Anthropic's Claude. A GSA employee confirmed to WIRED that the chatbot has been specifically adapted for secure use within the government. The DOGE team hopes to eventually use GSAi to analyze contract and procurement data, as WIRED previously reported.
The introduction of the chatbot raises questions about the overall strategy. A prominent AI expert, who wishes to remain anonymous, expressed concern to WIRED that providing AI tools to all employees could legitimize further layoffs.
In February, DOGE tested the chatbot in a pilot project with 150 users within the GSA. According to two sources familiar with the matter, DOGE hopes to eventually deploy the product across the entire agency. Development of the chatbot has been underway for several months, but the new, DOGE-aligned agency leadership has significantly accelerated the rollout schedule.
Federal employees can now interact with GSAi through a ChatGPT-like interface. The default model is Claude Haiku 3.5, but users can also choose Claude Sonnet 3.5 v2 and Meta LLaMa 3.2 depending on the task.
An internal memo about the product describes GSAi's various applications, including composing emails, creating talking points, summarizing texts, and writing code. At the same time, the memo warns against entering confidential federal government information or personal data as input. Another memo instructs users not to input data classified as "Controlled Unclassified Information."
The memo also includes instructions on writing effective prompts. An example shows how an imprecise request ("show newsletter ideas") can be refined ("I am planning a newsletter about sustainable architecture. Suggest 10 compelling topics related to environmentally friendly architecture, renewable energy, and reducing carbon footprint").
An employee who has already used the product describes GSAi's responses as "generic and predictable," comparable to the performance of an intern.
According to information obtained by WIRED, the Treasury Department and the Department of Health and Human Services have also recently explored the use of a GSA chatbot internally and in their external contact centers. Whether this chatbot is GSAi is unknown. In other areas of government, the US Army is already using a generative AI tool called CamoGPT to remove references to diversity, equity, inclusion, and accessibility from training materials, as WIRED previously reported.
In February, a project between the GSA and the Department of Education began to introduce a chatbot for support purposes within the Department of Education. The technical lead was DOGE employee Ethan Shaotran. In internal messages viewed by WIRED, GSA engineers discussed setting up a public "endpoint" – a specific access point on their servers – that would allow Department of Education officials to query an early pre-release version of GSAi. One employee described the setup as "makeshift" in a conversation with colleagues. The project was eventually discontinued, according to documents reviewed by WIRED.
In a staff meeting on Thursday, Thomas Shedd, a former Tesla engineer who now leads the Technology Transformation Services (TTS), announced that the GSA's tech department will shrink by 50 percent in the coming weeks, following the layoff of approximately 90 technology experts last week. Shedd plans for the remaining employees to work on more public-facing projects like Login.gov and Cloud.gov, which provide a variety of web infrastructure for other agencies. All other non-mandated work will likely be discontinued, according to Shedd.
Shedd has long supported the use of AI and automation in government. In early February, he told employees that he wanted to make AI a core part of the TTS agenda.