• Initial situation: An outdated Meta AI model was at the core of the Department of Government Efficiency's plan to purge parts of the federal government. Elon Musk's DOGE affiliates in the Office of Personnel Management used Meta's Llama 2 model to review and classify responses from federal workers to the "Fork in the Road" email.
• Related events: The "Fork in the Road" memo seemed to copy one Musk sent to Twitter employees, giving workers a choice. It was rumored that DOGE was feeding government employee data into AI, and records indicated Llama 2 was used to sort responses and count resignations. Llama 2 was also part of a scandal when Chinese researchers used it as the foundation for a military AI model. Meta then reversed policies and made Llama available for US national security applications.
• Deployment and concerns: Meta's open-source models can be used by the government without explicit consent. It's hard to track where Llama 2 was deployed in the government, and it's unclear why DOGE relied on Llama 2 when Meta has newer models. DOGE deployed Llama 2 locally to avoid sending data over the Internet, which was a privacy concern.
• Lawmaker's actions: More than 40 lawmakers sent a letter to the director of the Office of Management and Budget demanding a probe into DOGE's AI use due to serious security risks and potential to undermine AI adoption. The letter called out specific uses of AI models and emphasized concerns about security.
• Comparison with Grok: Grok was not available for DOGE's task in January as it was a proprietary model. Moving forward, DOGE may rely more on Grok as Microsoft will start hosting xAI's Grok 3 models. Lawmakers urged investigation of Musk's conflicts of interest and want more vetting of AI systems.
• Final stance: There is a fear that Musk may benefit from government data with his own models while putting it at risk. Lawmakers hope DOGE will unplug its AI systems, but the director seems more inclined towards innovation, while emphasizing the need for security and oversight.