#

Pentagon Breaks New Ground, AI Agents Now Authorized to Work With Classified Military Documents​​​​​​​​​​​​​​​​

Illustrative image generated by GPT
Illustrative image generated by GPT

Pentagon has cleared AI agents for classified documents, having previously limited them to unclassified material

The U.S. Department of War's Chief Digital and Artificial Intelligence Office (CDAO) has announced that AI agents from eight leading companies have been granted access to classified documents.

This comes after more than a month of agentic AI being used across the Pentagon, deployed at scale to work with reports, financial documentation, strategic documents, and similar materials, subject to human oversight and verification of final outputs, and restricted to unclassified access only.

Read more: Dutch Join U.S. Loyal Wingman Program With Two-Prototype Deal as CCA Competition Nears Decision

That restriction has now been lifted. AI agents will be used on classified-level documents and tasks. Notably, while the initial announcement listed seven companies, the list was expanded to eight within a matter of hours.

The companies that have signed agreements for AI agent deployment at the new classification level are:

  • SpaceX
  • OpenAI
  • Google
  • Nvidia
  • Reflection
  • Microsoft
  • Amazon Web Services
  • Oracle

Among the best-known consumer-facing products from these companies are ChatGPT, Gemini, and Copilot. Other firms on the list offer enterprise-grade solutions, while Nvidia is primarily a leading developer of AI hardware.

It is worth noting that agent-based AI interacts directly with systems through software interfaces (via APIs). It can autonomously execute complex tasks.

For example, with traditional generative AI, a human needs to provide sequential instructions and interact with it continuously. In contrast, an agent-based system can be given a final goal. It will then autonomously carry out the necessary intermediate tasks and processes step by step, learning in parallel.

At the same time, it remains unclear which specific operations the United States Department of War will involve these new artificial intelligence algorithms in, and this is unlikely to be disclosed due to their classified nature. It is also worth adding that the Department of War has already explicitly stated that there is no time for a cautious approach when it comes to integrating artificial intelligence algorithms into its activities.

Read more: Is UK's Skyhammer Anti-Shahed Interceptor Tested in Jordan Competitor to Ukraine's Anti-Aircraft Drones