Training versions of AI models on classified data is predicted to make them more accurate and effective in certain tasks, in line with a US defense official who spoke on background with . The news comes as demand for more powerful models is high: The Pentagon has reached agreements with OpenAI and Elon Musk’s xAI to operate their models in classified settings and is implementing a brand new agenda to develop into an “an ‘AI-first’ warfighting force” because the conflict with Iran escalates. (The Pentagon didn’t comment on its AI training plans as of publication time.)
Training could be done in a secure data center that’s accredited to host classified government projects, and where a replica of an AI model is paired with classified data, in line with two people conversant in how such operations work. Though the Department of Defense would remain the owner of the information, personnel from AI firms might in rare cases access the information in the event that they have appropriate security clearance, the official said.
Before allowing this latest training, though, the official said, the Pentagon intends to judge how accurate and effective models are when trained on nonclassified data, like commercially available satellite imagery.
The military has long used computer vision models, an older type of AI, to discover objects in images and pictures it collects from drones and airplanes, and federal agencies have awarded contracts to firms to coach AI models on such content. And AI firms constructing large language models (LLMs) and chatbots have created versions of their models fine-tuned for presidency work, like Anthropic’s Claude Gov, that are designed to operate across more languages and in secure environments. However the official’s comments are the primary indication that AI firms constructing LLMs, like OpenAI and xAI, could train government-specific versions of their models directly on classified data.
Aalok Mehta, who directs the Wadhwani AI Center on the Center for Strategic and International Studies and previously led AI policy efforts at Google and OpenAI, says training on classified data, as opposed to simply answering questions on it, would present latest risks.
