breaking March 2, 2026

OpenAI Amends Pentagon Contract Amid Employee Backlash

OpenAI amends its Pentagon contract banning domestic surveillance, as 500 AI employees back Anthropic’s military red lines on autonomous weapons.

OpenAI amends Pentagon contract amid employee backlash, a swift and unusual pressure campaign that came not just from the public, but from workers across the AI industry itself. CEO Sam Altman acknowledged the deal was rushed, saying it looked “opportunistic and sloppy,” and posted an internal memo to X announcing that the company is actively working with the Department of Defense to amend the agreement’s language on surveillance and autonomous weapons.

The contract amendments, which Altman described as clarifications to “make our principles very clear,” come after the original deal drew fierce scrutiny for potentially enabling domestic surveillance of U.S. citizens, a charge OpenAI’s own critics backed up by pointing to loopholes in the agreement’s references to Executive Order 12333.

How the Deal Fell Apart and Got Picked Up

The sequence of events began last Friday when the Pentagon’s negotiations with Anthropic collapsed. Secretary of Defense Pete Hegseth designated Anthropic a supply-chain risk and directed federal agencies to begin a six-month transition away from the company’s technology. The designation came after Anthropic drew hard red lines: its models would not be used in fully autonomous weapons systems or mass domestic surveillance programs, full stop.

Within hours of the Anthropic blacklisting, OpenAI announced its own deal to deploy AI models inside classified Pentagon networks. Altman said OpenAI maintained the same red lines as Anthropic, banning autonomous weapons and mass surveillance, but had found a path to a deal through a “multi-layered” safety architecture that included retaining full discretion over its safety stack, cloud-only deployment, and cleared OpenAI personnel remaining in the loop.

Critics immediately questioned the arrangement. Tech policy analyst Mike Masnick argued that the contract’s references to Executive Order 12333 effectively allowed the NSA to capture communications of U.S. persons through international intercepts, precisely the kind of domestic surveillance OpenAI claimed to prohibit.

The Employee Open Letter That Changed the Equation

The sharpest pressure on OpenAI, however, did not come from regulators or critics. By Monday, nearly 500 employees across OpenAI and Google had signed an open letter expressing support for Anthropic’s refusal to compromise on military AI red lines. The letter backed the principle that AI companies should maintain hard limits on use cases like mass surveillance and lethal autonomous systems regardless of government pressure or contract value.

The letter represented a significant internal signal, particularly notable given that many of the signatories were employees of OpenAI itself, essentially siding with a competitor’s ethics stance over their own employer’s deal.

The pressure appeared to work. Altman published his internal memo publicly on X, announcing that the Pentagon contract would be amended to include explicit language: “Consistent with applicable laws, including the Fourth Amendment to the United States Constitution, National Security Act of 1947, FISA Act of 1978, the AI system shall not be intentionally used for domestic surveillance of US persons and nationals.”

Altman also confirmed that intelligence agencies such as the NSA would be explicitly excluded from the current agreement scope, with any future use by those agencies requiring a separate contract modification.

What This Means for AI Company Strategy

The episode has crystallized a fault line that has been forming inside the AI industry for months: how far AI labs should go in cooperating with government and military clients, and under what conditions.

See also: The AI Safety Divide: Capabilities vs. Alignment | X01.

For related context, see GPT-5.4: Native Computer Use Now Beats Human Benchmarks.

The letter represented a significant internal signal, particularly notable given that many of the signatories were employees of OpenAI itself, essentially siding with a competitor’s ethics stance over their own employer’s deal.

The pressure appeared to work. Altman published his internal memo publicly on X, announcing that the Pentagon contract would be amended to include explicit language: “Consistent with applicable laws, including the Fourth Amendment to the United States Constitution, National Security Act of 1947, FISA Act of 1978, the AI system shall not be intentionally used for domestic surveillance of US persons and nationals.”

Altman also confirmed that intelligence agencies such as the NSA would be explicitly excluded from the current agreement scope, with any future use by those agencies requiring a separate contract modification.

What This Means for AI Company Strategy

The episode has crystallized a fault line that has been forming inside the AI industry for months: how far AI labs should go in cooperating with government and military clients, and under what conditions.

Anthropic’s position: draw hard red lines and accept being blacklisted rather than compromise. That stance has now attracted broad cross-industry support. Claude’s app surged to the top of Apple’s App Store download charts in the days following the designation, suggesting that consumers are paying attention to AI companies’ stated ethics positions and rewarding perceived principled behavior.

OpenAI’s position remains more complex. Altman has argued that engaging with government is better than ceding that ground to less safety-conscious vendors. The original blog post OpenAI published to defend the deal took an implicit shot at competitors, noting that other AI companies had “reduced or removed their safety guardrails” in national security deployments. But the rushed nature of the deal, signed the same day Anthropic was frozen out, undercut that narrative and lent credibility to the “opportunistic” framing.

The contract amendments, if finalized, would bring OpenAI’s formal commitments closer to Anthropic’s stated red lines. Whether they are enforceable in practice remains an open question, and one that the AI industry’s employees, regulators, and consumers are now actively pressing.

The broader implications are significant. For AI labs negotiating government contracts in 2026, this episode establishes that employee pressure and public backlash can materially reshape deal terms even after signing. For policymakers, it underscores the absence of any binding legal framework governing how AI systems can be used in classified military environments, a gap that advocates on both sides of the debate now say must be filled.

Related coverage: OpenAI Signs Pentagon Deal for Classified AI Deployment | Anthropic’s Blacklisting Triggers Backlash and App Store Surge