Elon Musk’s Department of Government Efficiency (DOGE), also known as DOGE for short, decided to shake things up by utilizing artificial intelligence from Meta’s Llama model to sift through and analyze emails from federal workers. The materials that were seen by WIRED reveal that DOGE members within the Office of Personnel Management (OPM) put Meta’s Llama 2 model to the test by reviewing and categorizing responses from federal workers to the controversial “Fork in the Road” email that circulated across the government in late January.
The infamous email presented federal employees with a choice: either resign if they disagreed with the Trump administration’s proposed changes to the federal workforce, such as a mandatory return-to-office policy, downsizing, and a demand for “loyalty,” or stay on board. All they had to do to leave their position was reply with the word “resign.” Interestingly, this email bore a striking resemblance to one that Musk himself sent to Twitter employees shortly after assuming control of the company in 2022. Records indicate that Llama was tasked with sorting through the email responses from federal workers to determine how many opted to accept the offer. From the materials viewed by WIRED, it seems that the model operated locally, reducing the likelihood of data being transmitted over the internet.
Despite numerous requests for comments from WIRED, both Meta and OPM remained silent on the matter. Mark Zuckerberg, the CEO of Meta, had made an appearance alongside other tech bigwigs such as Musk and Jeff Bezos at Trump’s inauguration in January. Surprisingly, not much information had been made public regarding his company’s technology being utilized within the government. Due to the open-source nature of Llama, government entities could easily leverage the tool to align with Musk’s objectives without requiring explicit approval from Meta. As the new administration under Trump took charge in January, DOGE operatives infiltrated OPM, an independent agency essentially serving as the federal government’s HR department. Their initial focus involved establishing a government-wide email service, as revealed by current and former OPM employees.
In late February, following the “Fork in the Road” email, OPM dispatched another communication to all government employees, prompting them to submit weekly summaries of their accomplishments in five bullet points. This initiative threw several agencies into disarray, with employees uncertain about managing email responses that needed to be cautious about security clearances and sensitive data. To add to the confusion, reports surfaced indicating that some workers who activated read receipts discovered that their responses weren’t actually being read. NBC News reported in February that these emails were slated for analysis through an AI system. While the materials reviewed by WIRED did not explicitly demonstrate DOGE affiliates utilizing Meta’s Llama models to analyze these weekly updates, it wouldn’t be a stretch for them to do so, according to insights from two federal workers. “We’re not entirely sure,” one federal worker remarked regarding DOGE’s potential use of Meta’s Llama for the “five points” emails. “But if they were clever, they’d recycle their code.”
