DOGE affiliates working inside the Office of Personnel Management (OPM) used Meta’s Llama 2 model to review and classify emails from federal workers, according to materials viewed by WIRED. The emails were responses to the government-wide “Fork in the Road” message sent in late January.
The use of Llama 2 shows how artificial intelligence was folded into a major workforce operation early in the Trump administration. The available records point to a practical goal: sorting replies and determining how many workers accepted a deferred resignation offer.
What DOGE Used Llama 2 to Review
The emails at the center of the review were responses to the “Fork in the Road” message. That message offered deferred resignation to federal workers who objected to changes the Trump administration was making to the federal workforce.
Those changes included an enforced return-to-office policy, downsizing, and a requirement to be “loyal.” Workers who wanted to leave their positions were told they only needed to reply with the word “resign.”
According to records described by WIRED, Llama was used to sort through those replies and classify them. The purpose was to identify how many workers accepted the offer.
The model appears to have run locally, based on the materials WIRED reviewed. That matters because a local deployment would make it unlikely that the email data was sent over the internet.
Why Meta’s Model Was Available
The AI system used in the review was not described as a custom government model. It was Meta’s Llama 2, part of the Llama family of models.
Meta CEO Mark Zuckerberg appeared with other Silicon Valley tech leaders, including Musk and Amazon founder Jeff Bezos, at Trump’s inauguration in January. But the WIRED report says little had been publicly known about Meta’s technology being used in government.
Llama’s open-source nature is central to the story. Because it can be used without the same kind of direct vendor arrangement required by some proprietary systems, the government could deploy it in support of Musk’s goals without Meta’s explicit consent.
Meta and OPM did not respond to WIRED’s requests for comment.
How OPM Became the Email Hub
OPM is an independent agency that functions as the human resources department for the federal government. Soon after Trump took office in January, DOGE operatives moved into the agency.
Current and former OPM employees told WIRED that the new administration’s first major goal for OPM was to create a government-wide email service. That service was used to send the original “Fork in the Road” email.
Riccardo Biasini, a former Tesla engineer, was involved in building the infrastructure for that email system, according to material viewed by WIRED and reviewed by two government tech workers.
The “Fork in the Road” message also echoed an earlier Musk playbook. WIRED reported that it closely mirrored an email Musk sent to Twitter employees shortly after he took over the company in 2022.
The Later “Five Points” Emails
Weeks after the “Fork in the Road” email, OPM sent another request to all government workers in late February. This one asked them to submit five bullet points describing what they accomplished each week.
Those emails created confusion across agencies. Workers were unsure how to handle responses that had to account for security clearances and sensitive information.
WIRED notes that some workers who turned on read receipts said they found that the responses were not actually being opened. NBC News reported in February that the emails were expected to go into an AI system for analysis.
The materials reviewed by WIRED do not explicitly show DOGE affiliates using Meta’s Llama models to analyze the weekly “five points” emails in the same way they used them for the “Fork in the Road” replies. Still, two federal workers told WIRED it would not be difficult to do so.
“We don’t know for sure,” says one federal worker on whether DOGE used Meta’s Llama to review the “five points” emails. “Though if they were smart they’d reuse their code.”
Where This Fits in DOGE’s AI Push
The Llama 2 deployment was one example of a broader pattern described by WIRED: DOGE has rolled out and used several AI-based tools across government agencies over the past few months.
WIRED reported in March that the US Army was using a tool called CamoGPT to remove DEI-related language from training materials. The General Services Administration also rolled out “GSAi” earlier this year, described as a chatbot intended to improve overall agency productivity.
OPM has also accessed software called AutoRIF, which could assist in the mass firing of federal workers.
One notable detail is what DOGE apparently did not use at first: Musk’s own AI model, Grok. When DOGE began building the government-wide email system in the first few weeks of the Trump administration, Grok was a proprietary xAI model and access to its API was limited.
That availability picture may change. Microsoft announced earlier this week that it would begin hosting xAi’s Grok 3 models as options in Azure AI Foundry, making those models more accessible in Microsoft environments like the one used at OPM. In February, Palantir struck a deal to include Grok as an AI option in its software, which is frequently used in government.
For now, the clearest documented case is Llama 2 and the “Fork in the Road” replies. It shows a government workforce email operation paired with an AI model to classify worker responses at scale, inside an agency already central to federal personnel decisions.