TLDR
- The Pentagon is pushing OpenAI, Anthropic, Google, and xAI to deploy AI tools on classified military networks without standard user restrictions
- OpenAI reached a deal this week to provide ChatGPT on genai.mil, an unclassified network serving 3 million Defense Department employees
- Only Anthropic’s Claude is currently available in classified settings through third parties, but the government must still follow the company’s usage policies
- OpenAI joined a $100 million Pentagon drone swarm program to develop voice-controlled software that translates verbal commands into digital instructions
- AI companies express concerns about removing safeguards, as mistakes in classified military settings could have deadly consequences
The Pentagon is in active negotiations with major AI companies to make their tools available on classified military networks. Pentagon Chief Technology Officer Emil Michael told tech executives at a White House event Tuesday that the military wants AI models on both unclassified and classified systems.
The talks involve OpenAI, Anthropic, Google, and xAI. A Pentagon official confirmed the department is “moving to deploy frontier AI capabilities across all classification levels.”
Currently, most AI tools for the military operate only on unclassified networks used for administrative work. Anthropic’s Claude is the only AI system available in classified settings through third parties.
Classified networks handle sensitive military operations including mission planning and weapons targeting. The Pentagon has not disclosed specific timelines for deploying AI chatbots on these systems.
Military officials want to use AI to process information and support decision-making. However, AI systems can make errors and generate false information that appears credible at first glance.
OpenAI Strikes Deal for Unclassified Network
OpenAI reached an agreement this week to provide its tools on genai.mil, an unclassified network. The system now serves more than 3 million Defense Department employees.
OpenAI removed many typical user restrictions as part of the deal. Some safeguards remain in place. An OpenAI spokesperson said expanding to classified networks would require a new or modified agreement.
Google and xAI have previously struck similar deals with the Pentagon. These agreements allow their AI tools to operate on military systems.
AI companies typically build safeguards into their models and require customers to follow usage guidelines. Pentagon officials have resisted these restrictions, arguing they should deploy commercial AI tools as long as they comply with American law.
Anthropic Negotiations Prove Contentious
Discussions between Anthropic and the Pentagon have been more difficult. Anthropic executives told military officials they do not want their technology used for autonomous weapons targeting or domestic surveillance.
“Anthropic is committed to protecting America’s lead in AI and helping the U.S. government counter foreign threats,” an Anthropic spokesperson said. The company stated Claude is already used for national security missions by the U.S. government.
The spokesperson said Anthropic is in productive discussions with the Department of War about continuing that work. President Donald Trump ordered the Department of Defense to rename itself the Department of War, a change requiring Congressional action.
OpenAI Joins Drone Swarm Program
OpenAI has partnered with two defense technology firms selected for a Pentagon drone swarm initiative. The $100 million “prize challenge” aims to develop voice-controlled software for autonomous drone fleets.
OpenAI’s technology will translate commanders’ verbal orders into digital instructions. The six-month competition seeks prototypes that can coordinate large autonomous systems through plain-language commands.
LtGen Frank Donovan, Director of the Defense Autonomous Warfare Group, said the Pentagon wants “orchestrator technologies that allow humans to work the way they already command – through plain language.” The challenge announcement came in January.
OpenAI’s role in the drone program is limited. The company’s tools will not be used for weapons integration, targeting authority, or direct drone operation.
An OpenAI spokesperson told Bloomberg the company did not submit its own bid for the prize. The company said its participation in the challenge will be cursory.


