OpenAI and the Pentagon Reach AI Agreement

New deal allows AI models in classified U.S. defense systems — with safety limits

OpenAI — the company that makes ChatGPT — has reached a major agreement with the U.S. Department of Defense to let its artificial intelligence models be used on classified military networks. The deal was announced on February 28, 2026, by OpenAI CEO Sam Altman and comes amid a bitter dispute between the Pentagon and another AI company, Anthropic.

This move marks a significant moment in how the U.S. government uses AI technology in national security, and it shows how safety and military needs are being balanced in real world applications.

What the Agreement Says

Under the agreement:

• OpenAI will deploy its AI models inside classified networks used by the U.S. military.
• The company will keep control of its own safety systems and how models are used.
• The Pentagon agreed to avoid using OpenAI systems for domestic mass surveillance or fully autonomous weapons — two key “red lines” OpenAI has said are important.
• OpenAI’s engineers will work with the Defense Department to help ensure the AI behaves safely and as expected.

Altman said that the Department of Defense “displayed a deep respect for safety” and agreed to terms that match OpenAI’s principles. He also said OpenAI hopes other AI companies can reach similar agreements.

Why This Matters

The deal is important because it shows a midpoint between AI safety concerns and national defense needs:

• The Pentagon wants top-level AI that can help with planning, data analysis, logistics, cybersecurity, and other military tasks.
• At the same time, many experts and companies are cautious about letting AI control weapons or participate in mass surveillance.
• OpenAI’s agreement limits the use of its technology so it won’t be used in ways the company considers dangerous or unethical.

This kind of arrangement — where a private AI company still controls how its models are used — is relatively new and may shape future tech–government partnerships.

Conflict With Anthropic and Pentagon

This agreement comes right after a public conflict between the Pentagon and Anthropic, another major AI company. The Pentagon demanded that Anthropic lift certain safety limits on how its AI could be used in defense projects. Anthropic refused, saying it could not in “good conscience” allow its AI to be used for some military applications, especially autonomous weapons and domestic surveillance.

In response, the Pentagon — backed by President Donald Trump — moved to stop all federal agencies from using Anthropic’s AI. Some Pentagon leaders even called Anthropic a national security risk and warned it could be banned from future contracts.

OpenAI’s deal followed hours after these actions, leaving many to see it as part of a broader shift in which AI companies are accepted for defense work.

OpenAI’s Safety Approach

In his public statement and internal company messages before the deal was finalised, Altman emphasised that OpenAI wants AI to be safe and widely beneficial, even in military contexts. He noted that the Pentagon deal allows OpenAI to keep “technical safeguards” in place — systems that sit between a powerful AI model and military use, so the model can refuse dangerous tasks and remain under human control.

Part of this approach includes deploying AI only in cloud systems, not directly inside autonomous weapons or aircraft. If the AI refuses to perform a harmful task, the U.S. government “will not force it to comply,” Altman said.

What Happens Next

• Officials say the deal does not give OpenAI total control — the Department of Defense still decides how and when the AI models are used.
• Other tech firms may seek similar agreements, though the situation with Anthropic shows how disagreements over ethical limits can complicate government partnerships.
• Congress and the public will continue to debate the role of AI in military and national security policy.

Analysts say this deal could become a template for how AI companies work with government agencies — balancing innovation, safety, and national interests.

In Short

• OpenAI reached a deal with the Pentagon to use its AI in classified defense networks.
• The company kept ethical safeguards against surveillance and autonomous weapons.
• The agreement follows a fight between the Pentagon and Anthropic, which refused to remove its own safety limits.
• Experts say this reflects the challenge of combining AI technology with national security priorities.