On February 28, OpenAI finalized a deal to supply its AI technologies to the US military for classified operations, hours after the Pentagon banned Anthropic for refusing to comply with its demands.
In this DotNXT Tech story, we examine how OpenAI’s legalistic approach is forcing a reckoning across the AI industry.
Why the Deal Happened Now
The Pentagon’s ultimatum to Anthropic was the catalyst. After Anthropic refused to drop its contractual prohibitions on autonomous weapons and mass surveillance, Defense Secretary Pete Hegseth labeled the company a “supply chain risk” and barred federal contractors from working with it. OpenAI, sensing opportunity, rushed negotiations that Altman later called “definitely rushed.”
The timing was no accident. The Pentagon launched strikes on Iran the same night the ban took effect, and Hegseth gave the military six months to replace Anthropic’s Claude with OpenAI’s models and xAI’s systems. The message was clear: compliance or obsolescence.
OpenAI’s gamble paid off. It won the contract while Anthropic faces a scorched-earth campaign that could cripple its business.
OpenAI’s Legalistic Approach vs. Anthropic’s Moral Stand
OpenAI’s contract relies on existing laws—like the 2023 Pentagon directive on autonomous weapons and the Fourth Amendment—to set boundaries. Altman argued this was more practical than Anthropic’s “specific prohibitions,” which the Pentagon rejected as overreach. The company’s blog post framed the deal as a victory for both business and ethics.
But the legal safeguards are porous. OpenAI’s published contract excerpt admits it has no “free-standing right” to block lawful military uses. Jessica Tillipman, a government procurement law expert, noted the agreement merely restates that the Pentagon can’t break current laws—a low bar given AI’s potential to expand surveillance under existing rules.
Anthropic’s stance, though unsuccessful, exposed the flaw in OpenAI’s logic. If the government’s track record on surveillance (see: Snowden) is any guide, legal compliance is not a reliable safeguard. OpenAI’s head of national security partnerships argued that if you distrust the government’s adherence to law, you should also distrust its adherence to contractual red lines. That’s a false equivalence. Contracts create enforceable obligations; laws are often reinterpreted to fit political needs.
DotNXT’s Take: OpenAI’s deal is less about safety than about survival. The company is betting that legalistic wiggle room will placate both the Pentagon and its employees. It’s a high-stakes gamble that could backfire if the military pushes the boundaries of “lawful” use.
Safety Controls: Real Protection or PR?
OpenAI claims it will embed “red lines” directly into its models to prevent mass surveillance and autonomous weapons use. Boaz Barak, an OpenAI employee, wrote on X that the company’s safety rules will apply even in classified settings. But the company hasn’t explained how these rules differ from its standard user protections, nor how it will enforce them in a six-month rollout.
Enforcement in classified environments is inherently opaque. OpenAI’s contract excerpt is vague on oversight mechanisms, and the company has not responded to requests for clarification. The Pentagon’s urgency to deploy AI in Iran and Venezuela suggests it won’t tolerate delays, even for safety checks.
The bigger question is whether tech companies should be the arbiters of military ethics. The Pentagon’s Hegseth made it clear: the government views contractual prohibitions as unacceptable interference. OpenAI’s deal sidesteps this by deferring to the law, but that deference may come at the cost of meaningful oversight.
Fallout for Anthropic and the AI Industry
Anthropic’s refusal to bend cost it dearly. The Pentagon’s ban extends beyond its own contracts—any company doing business with the military is now barred from working with Anthropic. The company has vowed to sue, but legal experts question whether the government can legally enforce such a broad restriction.
OpenAI, meanwhile, has positioned itself as the Pentagon’s preferred AI vendor. The deal includes a six-month phase-out of Claude, which was reportedly used in the Iran strikes hours after the ban. The transition won’t be seamless. The military’s reliance on Claude for classified operations suggests OpenAI’s models will face immediate pressure to perform in high-stakes scenarios.
The industry is watching closely. If OpenAI’s legalistic approach becomes the norm, other AI companies may abandon moral stands in favor of pragmatism. The alternative—being locked out of the world’s largest military market—is a risk few can afford.
FAQ
What does OpenAI’s Pentagon deal actually allow?
The contract permits the US military to use OpenAI’s technologies in classified settings, but with two stated prohibitions: no mass domestic surveillance and no use in autonomous weapons without human involvement. However, these prohibitions are not contractual guarantees. OpenAI’s agreement relies on existing laws, which critics argue are too permissive to prevent misuse. The company has not disclosed how it will enforce its “red lines” in classified environments.
How is OpenAI’s approach different from Anthropic’s?
Anthropic sought explicit contractual prohibitions on autonomous weapons and mass surveillance, which the Pentagon rejected as unacceptable interference. OpenAI, by contrast, framed its safeguards as compliance with existing laws, such as the 2023 Pentagon directive on autonomous weapons and the Fourth Amendment. This legalistic approach allowed OpenAI to secure the deal, but it provides weaker protections than Anthropic’s proposed terms.
Why did the Pentagon ban Anthropic?
The Pentagon banned Anthropic after the company refused to drop its contractual prohibitions on autonomous weapons and mass surveillance. Defense Secretary Pete Hegseth accused Anthropic of “arrogance and betrayal” and declared it a “supply chain risk.” The ban extends beyond the Pentagon’s own contracts—any company doing business with the military is now prohibited from working with Anthropic.
What are the risks of OpenAI’s deal?
The primary risk is that OpenAI’s reliance on legal safeguards will prove insufficient. The company’s contract does not grant it the right to block lawful military uses, and enforcement in classified settings is opaque. Critics warn that the deal could enable the expansion of surveillance and autonomous weapons under the guise of compliance with existing laws. There’s also the risk of employee backlash—OpenAI’s workforce has historically been vocal about ethical concerns.
What happens next for Anthropic?
Anthropic faces an existential threat. The Pentagon’s ban could cripple its business if enforced, as it bars any company with military contracts from working with Anthropic. The company has vowed to sue, but the legal battle will be uphill. In the meantime, the military is phasing out Anthropic’s Claude model, which was reportedly used in recent strikes on Iran.
How will this deal affect the AI industry?
The deal sets a precedent that could reshape the AI industry’s relationship with the military. OpenAI’s legalistic approach may become the template for future contracts, as companies prioritize market access over moral stands. The Pentagon’s aggressive stance against Anthropic sends a clear message: non-compliance will not be tolerated. Smaller AI firms may now feel pressured to abandon ethical red lines to avoid being locked out of lucrative defense contracts.
Conclusion
OpenAI’s deal with the Pentagon is a calculated retreat from moral absolutism. The company has traded Anthropic’s principled stand for a seat at the table, betting that legalistic safeguards will hold. That bet may pay off in the short term, but it risks normalizing a dangerous precedent: that AI companies must defer to the military’s interpretation of the law. The real test will come when the Pentagon pushes the boundaries of “lawful” use—and whether OpenAI’s red lines hold or fold.
