Meta’s Llama AI Secretly Used to Analyze Federal Worker Emails Under Musk’s DOGE Program

DOGE Used Meta’s Llama 2 AI Model to Review Federal Employee Emails

The Department of Government Efficiency (DOGE), affiliated with Elon Musk, tested and used Meta’s Llama 2 AI model to analyze emails from federal workers. These emails were responses to the so-called “Fork in the Road” message sent across government agencies in late January.

The “Fork in the Road” email offered federal employees the option to resign if they opposed changes introduced by the Trump administration. These changes included a strict return-to-office policy, workforce downsizing, and a loyalty requirement. To resign, workers simply had to reply with the word “resign.” The email closely resembled one Musk sent to Twitter employees after acquiring the company in 2022.

How Llama 2 Was Used in the Federal Workforce

Documents reviewed by WIRED reveal that Llama 2 was deployed locally within the Office of Personnel Management (OPM) to classify and sort email responses. This local deployment likely prevented sensitive data from being sent over the internet. Neither Meta nor OPM responded to inquiries about these operations.

Meta’s open-source Llama 2 model allows government agencies to utilize its capabilities without direct involvement from the company. This means federal programs like DOGE could apply the AI to support administrative goals independently.

Background: DOGE’s Role and Email Initiatives

Shortly after Trump’s inauguration, DOGE operatives embedded within OPM, the federal government’s HR agency, prioritized creating a government-wide email system. Riccardo Biasini, a former Tesla engineer, helped build the infrastructure that delivered the original “Fork in the Road” email.

Following the “Fork” email, OPM sent another request in late February asking federal employees to submit weekly bullet points summarizing their accomplishments. This sparked confusion across agencies, as workers worried about security clearance and the handling of sensitive information. Some employees who enabled read receipts noticed their responses were not being opened.

AI Analysis of Weekly Reports and Future Potential

While it is not confirmed that DOGE used Llama 2 to review these weekly accomplishment emails, federal workers suggest it would be logical for the agency to reuse the existing AI tools. The infrastructure to do so is already in place.

Initially, DOGE did not use Elon Musk’s AI model, Grok, for the government-wide email system, as it was proprietary and access was limited. However, Microsoft recently announced it will host Grok 3 models on its Azure AI Foundry. This integration could make Grok more accessible within Microsoft environments like OPM’s, potentially enabling its future use.

Other AI Tools in Use Across Federal Agencies

• Palantir incorporated Grok as an AI option in its widely used government software.
• The US Army has used CamoGPT to remove diversity, equity, and inclusion (DEI) language from training materials.
• The General Services Administration (GSA) launched “GSAi,” a chatbot designed to improve agency productivity.
• OPM has accessed AutoRIF software, which could assist in large-scale employee terminations.

These examples show a growing integration of AI tools within government agencies to streamline operations and enforce policy changes.

Federal employees interested in AI applications and training can explore AI courses relevant to government roles to understand how these tools impact their work environment.