On February 27, 2026, within hours of each other, two events reshaped the AI industry. First, President Trump ordered all U.S. federal agencies to stop using Anthropic's products — with a six-month transition period. Then, OpenAI announced it had signed a contract with the Department of Defense to deploy its models on the Pentagon's classified networks. Estimated value: $200 million, potentially reaching $500 million to $2 billion over five years according to defense sector analysts.
This was no coincidence. It was an orchestrated sequence — and it sent a shockwave through the tech world that no one had anticipated at this scale.
What Anthropic Refused
To understand the OpenAI-Pentagon deal, you have to start with what came before it. Anthropic had been in advanced negotiations with the Department of Defense for a similar classified AI deployment contract. The talks broke down over two specific points: Anthropic demanded a contractual ban on using its Claude models for mass surveillance of American citizens, and a ban on integrating them into autonomous weapons systems.
The Pentagon rejected those restrictions. Anthropic held its red lines. In response, Secretary of Defense Pete Hegseth designated Anthropic as a "national security supply chain risk" — a label normally reserved for hostile foreign companies. Anthropic threatened to challenge the designation in court.
Within hours of this breakdown, OpenAI was ready to sign.
OpenAI's "Red Lines" — and Their Loopholes
OpenAI presented its deal with three main guardrails: no domestic mass surveillance, no directing autonomous weapons systems, no high-stakes automated decisions (such as "social credit" scoring). The company retains control of its security stack and deploys via the cloud, with cleared personnel in the loop.
On paper, it looks like the same lines Anthropic drew. In practice, the difference lies in the contractual fine print — and especially in what was never made public. The full text of the contract has never been disclosed. The Electronic Frontier Foundation flagged the phrase "consistent with applicable laws," noting that the U.S. government has historically interpreted this clause extremely broadly to justify surveillance programs — as Edward Snowden revealed in 2013.
Another gray area: the original contract did not explicitly exclude intelligence agencies from its scope. It took an amendment, signed on March 2, 2026, after a public outcry, to specify that the NSA and other intelligence agencies within the Department of Defense were excluded from this specific agreement.
The word "specific" is doing all the heavy lifting in that sentence.
"Opportunistic and Sloppy"
Sam Altman himself acknowledged the problem. On March 3, facing the scale of the backlash, the OpenAI CEO stated in interviews reported by CNBC and Bloomberg: "We were genuinely trying to defuse the situation and avoid a worse scenario, but I think it just came across as opportunistic and sloppy." He admitted he should not have rushed the announcement on a Friday evening.
This admission is remarkable. Not so much for its content — the opportunistic timing was obvious to everyone — but for what it reveals about internal pressure. Because the backlash wasn't only coming from outside.
Several OpenAI employees signed an open letter supporting Anthropic. On March 7, Caitlin Kalinowski, OpenAI's head of robotics and hardware, resigned. Her explanation, posted publicly: "Surveilling Americans without judicial oversight and autonomous lethality without human authorization are lines that deserved more deliberation than they got. This isn't primarily a technical question. It's a governance question."
Kalinowski was no recent hire. A veteran of Apple (six years on MacBooks), Oculus (nine years in VR), and Meta (two and a half years on the Orion AR glasses), she had been leading OpenAI's robotics operations since November 2024.
1.5 Million Walk Away
The #QuitGPT movement reached 1.5 million participants in the days following the announcement. Claude, Anthropic's chatbot, climbed to the number one spot on the U.S. App Store in under 24 hours, dethroning ChatGPT.
This shift illustrates a new phenomenon: a significant share of mainstream AI users now make product choices based on ethical criteria, not just features. You can debate whether this movement will last — tech boycotts have historically had short memories. But the signal it sends to investors and boards is real: signing with the military carries a measurable brand cost.
What This Means for Businesses
For businesses deploying or considering deploying AI, this episode isn't just an American affair. It raises three concrete questions.
The first is vendor dependency. Anthropic was banned from the U.S. government in 48 hours. Whoever your AI providers are, you're exposed to geopolitical decisions entirely beyond your control. It's one more reason not to lock your entire AI stack into a single vendor.
The second is data. If your data flows through models deployed on military infrastructure — even via the cloud — confidentiality takes on a different meaning. The Model Context Protocol (MCP), which standardizes connections between AI agents and enterprise systems, makes these integrations easier but also deeper. The more access an agent has to your systems, the more critical the question of who controls the underlying infrastructure becomes.
The third is reputation. Your customers, your recruits, your partners are starting to ask: "Which AI are you using, and why?" Today, it's a fringe concern. In two years, it could be a vendor selection criterion — much like GDPR compliance became one.
A Precedent, Not a One-Off
The OpenAI-Pentagon deal is not an isolated event. It's the logical outcome of the race to monetize generative AI. OpenAI is burning through cash at a rate that makes government contracts not just desirable but necessary. The U.S. defense market is the largest possible customer for a company that needs massive, predictable recurring revenue.
Anthropic made a different choice — and paid the immediate price. Whether that choice is sustainable long-term will depend on its ability to find equivalent revenue elsewhere. But at least the line is clear.
What should concern observers isn't that OpenAI signed. It's how fast it all happened. One vendor ousted, one contract signed, guardrails bolted on after the fact under public pressure, full text never published. All in less than a week. Decisions that affect the surveillance of hundreds of millions of people probably deserve better than a Friday evening and a press release.