OpenAI Strikes Pentagon Deal to Deploy AI on Classified Networks, With Safety Strings Attached | X01
Sam Altman announced a landmark agreement to bring OpenAI
breaking February 28, 2026
OpenAI Strikes Pentagon Deal to Deploy AI on Classified Networks, With Safety Strings Attached
Sam Altman announced a landmark agreement to bring OpenAI’s models into the U.S. Department of Defense’s classified cloud infrastructure, hours after the Trump administration banned rival Anthropic over a parallel standoff. The deal sets a new precedent for how commercial AI enters military systems, and on what terms.
OpenAI has secured a landmark deal to deploy its AI models inside the U.S. Department of Defense’s classified cloud networks, with explicit guardrails banning autonomous weapons and domestic mass surveillance. CEO Sam Altman announced the agreement late Friday, just hours after the Trump administration ordered federal agencies to cut ties with rival Anthropic following a very different kind of conversation about the same question.
The contrast is stark. Two major AI labs. Two approaches to the same demand. One is now inside the Pentagon’s classified systems. The other has been declared a national security risk.
The Deal
In a post on X, Altman confirmed that OpenAI reached an agreement with the U.S. Department of Defense to provide its AI technologies for classified systems. “Tonight, we reached an agreement with the Department of War to deploy our models in their classified network,” Altman wrote, using the recently reinstated name for the Defense Department.
Altman said the DoD “displayed a deep respect for safety and a desire to partner to achieve the best possible outcome.” The deal includes explicit commitments: OpenAI’s technology will not be used for domestic mass surveillance or for programming autonomous weapon systems, platforms capable of selecting and attacking targets without human intervention. Humans, Altman affirmed, will retain “responsibility for the use of force.”
“We remain committed to serve all of humanity as best we can,” Altman added. “The world is a complicated, messy, and sometimes dangerous place.”
The Anthropic Contrast
The OpenAI announcement landed against the backdrop of one of the more dramatic confrontations in recent AI history. Anthropic CEO Dario Amodei had told the Pentagon he could not “in good conscience accede” to certain demands, specifically requests to remove safety guardrails around autonomous weapons and domestic surveillance capabilities from Claude.
The White House responded by designating Anthropic a supply-chain risk to national security and ordering all federal agencies to stop using its products. The administration’s move was posted directly to social media before formal agency notification, an escalation that signaled how personally the confrontation had been taken.
Reports have also emerged that Claude had previously been deployed by U.S. military personnel to assist in operational missions, a use case Anthropic says was outside agreed parameters and that contributed to the company drawing harder lines on acceptable deployments.
What OpenAI Negotiated
The specific carve-outs Altman secured are not incidental. They mirror almost exactly the safeguards Anthropic said it refused to remove. The convergence is notable: both companies drew the same lines in the sand. The difference is that OpenAI appears to have gotten the Pentagon to accept those lines, at least in writing.
Whether those commitments hold in classified operational contexts is a different question, and one that neither the DoD nor OpenAI addressed publicly. Defense contracts of this nature are not subject to public audit. The guardrails exist as declared policy, and enforcement is a matter of trust and internal oversight.
Human rights organizations have already raised concerns. AI systems deployed in military contexts have a track record of scope creep. Earlier AI-assisted targeting tools used by other militaries offered an early blueprint for how military AI can be operationalized in ways that outrun the ethical frameworks built around them.
The Strategic Pivot
For OpenAI, this deal represents a significant shift. The company has historically positioned itself as a civilian-focused AI provider. Its previous usage policies explicitly excluded lethal autonomous weapons. That position now carries an asterisk: it will provide its models to classified military infrastructure, with contractual but unverifiable restrictions on specific applications.
The deal also sharpens OpenAI’s commercial position. With Anthropic effectively sidelined from the federal government’s AI stack, OpenAI steps into a substantial vacuum. Government contracts in this tier carry enormous value, financial, strategic, and reputational, for a company whose nonprofit-to-for-profit restructuring has attracted growing investor scrutiny over mission fidelity.
The Bigger Picture
The week ending February 28 has produced a crystallizing moment for AI governance in military contexts. One lab held its ground on safety and lost its government business. Another negotiated guardrails and won the contract. The argument that commercial AI labs can simultaneously serve defense clients and maintain meaningful ethical constraints has moved from theoretical to empirical, and OpenAI is now the test case.
For related context, see The Pentagon.
The White House responded by designating Anthropic a supply-chain risk to national security and ordering all federal agencies to stop using its products. The administration’s move was posted directly to social media before formal agency notification, an escalation that signaled how personally the confrontation had been taken.
Reports have also emerged that Claude had previously been deployed by U.S. military personnel to assist in operational missions, a use case Anthropic says was outside agreed parameters and that contributed to the company drawing harder lines on acceptable deployments.
What OpenAI Negotiated
The specific carve-outs Altman secured are not incidental. They mirror almost exactly the safeguards Anthropic said it refused to remove. The convergence is notable: both companies drew the same lines in the sand. The difference is that OpenAI appears to have gotten the Pentagon to accept those lines, at least in writing.
Whether those commitments hold in classified operational contexts is a different question, and one that neither the DoD nor OpenAI addressed publicly. Defense contracts of this nature are not subject to public audit. The guardrails exist as declared policy, and enforcement is a matter of trust and internal oversight.
Human rights organizations have already raised concerns. AI systems deployed in military contexts have a track record of scope creep. Earlier AI-assisted targeting tools used by other militaries offered an early blueprint for how military AI can be operationalized in ways that outrun the ethical frameworks built around them.
The Strategic Pivot
For OpenAI, this deal represents a significant shift. The company has historically positioned itself as a civilian-focused AI provider. Its previous usage policies explicitly excluded lethal autonomous weapons. That position now carries an asterisk: it will provide its models to classified military infrastructure, with contractual but unverifiable restrictions on specific applications.
The deal also sharpens OpenAI’s commercial position. With Anthropic effectively sidelined from the federal government’s AI stack, OpenAI steps into a substantial vacuum. Government contracts in this tier carry enormous value, financial, strategic, and reputational, for a company whose nonprofit-to-for-profit restructuring has attracted growing investor scrutiny over mission fidelity.
The Bigger Picture
The week ending February 28 has produced a crystallizing moment for AI governance in military contexts. One lab held its ground on safety and lost its government business. Another negotiated guardrails and won the contract. The argument that commercial AI labs can simultaneously serve defense clients and maintain meaningful ethical constraints has moved from theoretical to empirical, and OpenAI is now the test case.
The terms Altman secured will be scrutinized closely, and not just by competitors. Every AI safety researcher watching this deal will be asking the same question: when an AI model is running inside a classified network, who actually enforces the contract?