Training versions of AI models on classified data is expected to make them more accurate and effective at certain tasks, according to a US defense official who spoke on background. MIT Technology Review. The news comes as demand for more powerful models is high: The Pentagon has signed deals with OpenAI and Elon Musk’s XAI to operate its models in classified settings and is implementing a new one. agenda To become “an ‘AI-first’ fighting force” as conflict with Iran escalates. (The Pentagon had not commented on its AI training plans by the time of publication.)
The training will be conducted in a secure data center that is accredited to host classified government projects, and where a copy of the AI ​​model is paired with classified data, according to two people familiar with how such operations work. Although the Defense Department will remain the owner of the data, personnel at AI companies can access the data in rare cases if they have the appropriate security clearance, the official said.
However, before allowing this new training, the official said, the Pentagon intends to evaluate how accurate and effective the models are when trained on unclassified data, such as commercially available satellite imagery.
The military has long used computer vision models, an older form of AI, to identify objects in images and footage collected from drones and airplanes, and federal agencies contracts awarded Companies to train AI models on such content. And AI companies making large language models (LLMs) and chatbots have created versions of their models for government work, such as Anthropic’s Cloud Gov, designed to operate in more languages ​​and secure environments. But the official’s comments are the first sign that AI companies that create LLMs, like OpenAI and XAI, can train government-specific versions of their models directly on classified data.
Alok Mehta, who directs the Wadhwani AI Center at the Center for Strategic and International Studies and previously led AI policy efforts at Google and OpenAI, says training on classified data, rather than just answering questions about it, will introduce new risks.
