FDA Employees Told Not to Use Claude AI in Tech Dispute
HHS has directed FDA and other health agency employees to immediately stop using the artificial intelligence (AI) platform Claude, disabling access across its enterprise systems after the Trump administration labeled parent company Anthropic a “supply chain risk,” according to a post on NOTUS.org. In an email sent Monday by the HHS Office of the Chief Information Officer and reviewed by NOTUS, employees were told they would “no longer be able to log in to or access Claude through the HHS enterprise environment” and were instructed to transition ongoing work to other department-approved AI tools.
The move follows a breakdown in negotiations between Anthropic and the U.S. Department of Defense over terms governing the government’s use of the company’s systems. Trump, writing on Truth Social, accused Anthropic of attempting to “strong-arm” the Pentagon by insisting the department adhere to the company’s terms of service rather than federal directives. Anthropic CEO Dario Amodei said the company’s position stemmed from concerns about potential uses of its technology for mass domestic surveillance and fully autonomous weapons.
The directive could have significant implications for FDA, which has accelerated its adoption of artificial intelligence tools under the Trump administration in an effort to speed regulatory reviews. One FDA employee, speaking anonymously to discuss internal matters candidly, told NOTUS that Anthropic’s Claude Sonnet 4.5 had become deeply integrated into scientific review workflows, regulatory writing and statistical coding. “For serious scientific work, study reviews and regulatory writing or R-coding, there is no substitute,” the employee said, adding that staff had spent the past year refining prompts and training reviewers to use the system effectively. “This decision, if confirmed, will basically wipe out 18 months of efforts,” the employee said.
FDA last year launched its internal AI platform, Elsa, which provided access to Claude alongside other large language models. Although Claude access remained active for some users immediately after the HHS directive, an internal FDA notice indicated the platform would transition away from Anthropic models.
“Gemini is already available in Elsa and will become the primary model going forward,” the internal banner stated, referring to Google’s AI system. HHS confirmed that OpenAI’s ChatGPT and Google Gemini remain approved for enterprise use.
Still, some staff expressed skepticism about whether Gemini can match Claude’s performance for technical and analytical tasks. The FDA employee described Gemini as effective for conversational interactions but less suited for “in-depth analysis of complex scientific study reports,” according to NOTUS.