The U.S. Department of Defense is urging several leading artificial intelligence companies, including OpenAI and Anthropic, to connect generative AI tools widely used in commercial fields to military classified networks, and to remove or significantly relax regular use restrictions for ordinary users in these environments.
According to people familiar with the matter, at an event held at the White House on Tuesday, Defense Department Chief Technology Officer Emil Michael told multiple technology company executives that the military’s goal is to enable these cutting-edge AI models to run on both unclassified and classified information systems. An unnamed official said the Pentagon is "promoting the deployment of cutting-edge AI capabilities at all classified levels."
The move is part of ongoing negotiations between the Pentagon and several generative AI companies on how the United States will use artificial intelligence on future battlefields; in these scenarios, drone swarms, robots and cyberattacks have become part of the war picture. Michael's statement is expected to further intensify the sharp debate between "the military's desire to use AI with as few restrictions as possible" and "technology companies trying to set usage boundaries for their own tools."
Currently, many AI companies are developing customized tools for the U.S. military, but most of the products can only be deployed on unclassified networks used for daily military management. In a classified environment, only some of Anthropic's products are available through third-party channels, and governments are still required to abide by the company's usage policies. The military's classified networks are typically used to handle more sensitive tasks, such as combat operation planning or weapons targeting; it is unclear how and on what timeline the Pentagon plans to fully embed chatbot-like AI systems into these networks.
Military officials hope to use the powerful information integration and induction capabilities of generative AI to assist decision-making and situation analysis. But researchers warn that even the most advanced models can make errors and even "make up" information that sounds reasonable but is actually false; once such errors occur in a highly sensitive and classified combat environment, the consequences can be fatal.
In order to reduce risks as much as possible, most AI companies have built-in security protection mechanisms within the models and require customers to comply with a series of usage guidelines. However, some Pentagon officials have criticized these external restrictions, arguing that the military should have the right to deploy and use commercial AI tools according to its own needs as long as it complies with U.S. law.
This week, OpenAI reached an agreement with the Pentagon to allow the military to use its tools, including ChatGPT, on an unclassified network called "genai.mil"; the platform has been made available to more than 3 million Defense Department employees. As part of the agreement, OpenAI agreed to lift many of the network's usual restrictions on public use, although some safety guardrails will remain in place. Google, owned by Alphabet, and xAI, founded by Musk, have previously reached similar arrangements.
OpenAI said in a statement that this cooperation is currently only for non-confidential use through "genai.mil". To expand on this basis to other environments (such as confidential networks), a new or revised agreement will need to be signed.
Compared with OpenAI, Anthropic’s negotiation process with the Pentagon is more tense and tricky. Company executives have made it clear to military officials that they do not want their technology to be used for autonomous weapons targeting or large-scale surveillance in the United States. Anthropic's products include the chatbot "Claude."
An Anthropic spokesperson said the company is "committed to maintaining U.S. leadership in AI and helping the U.S. government respond to foreign threats by providing the warfighter with state-of-the-art AI capabilities." The spokesperson said that Claude has been widely used in multiple national security missions of the U.S. government and that the company is maintaining a "productive dialogue" with the "Department of War" on how to continue to advance related cooperation.
At the order of President Donald Trump, the U.S. Department of Defense is moving forward with plans to change its name to the Department of War, subject to legislative action by Congress.