OpenAI amended its US defence department contract days after signing it on 27 February 2026, after OpenAI’s chief executive, Sam Altman, admitted the rush “looked opportunistic and sloppy”.
The agreement, signed on Friday, put OpenAI’s models into classified military operations, landing hours after talks between Anthropic and defence secretary Pete Hegseth collapsed.
For those not in the know, Hegseth wanted AI to be able to kill Donald Trump’s enemies without human intervention and to be able to spy on US citizens. Anthropic said no, and Hesgeth said that he could not work with such a woke company. Which was when OpenAI stepped in and apparently claimed to offer the service the government wanted
OpenAI boasted its deal had “more guardrails than any previous agreement for classified AI deployments, including Anthropic’s”.
On 2 March 2026, Altman said OpenAI was adding terms so “the AI system shall not be intentionally used for domestic surveillance of US persons and nationals”.
He said intelligence services such as the National Security Agency would be excluded for now.
Altman’s issue was that his own staff were raising red flags about signing up for this deal, and the weekend brought chalk graffiti outside its San Francisco office reading “NO TO MASS SURVEILLANCE” and “Do the right thing!”
Anthropic and OpenAI have talked tough about surveillance and weapons without human oversight, which made it awkward when OpenAI got a deal that Anthropic could not.
OpenAI said it could hold its red lines with technical controls, including deploying only through the cloud and keeping employees in the loop.
Altman said on 28 February 2026: “Anthropic seemed more focused on specific prohibitions in the contract, rather than citing applicable laws, which we felt comfortable with.”
On Monday, he conceded the broader concern that AI can turbocharge mass data-gathering.
“We shouldn’t have rushed to get this out on Friday. The issues are super complex and demand clear communication. We were genuinely trying to de-escalate things and avoid a much worse outcome, but I think it just looked opportunistic and sloppy,” he wrote.
The new language would “prohibit deliberate tracking, surveillance or monitoring of US persons or nationals, including through the procurement or use of commercially acquired personal or identifiable information”.
Anthropic’s talks cratered after both sides failed to agree on wording, with Anthropic’s chief executive, Dario Amodei, drawing two red lines against domestic mass surveillance and lethal autonomous weapons.
Hegseth wanted the models available for “all lawful use”, while Anthropic argued US law still left space for mass surveillance using AI tools.
The Pentagon signalled on 27 February 2026 it could remove phrases Anthropic said were too vague, and senior figures thought a deal was close, according to a person with direct knowledge.
Then it fell apart anyway, with the crunch point being terms around mass collection of publicly available data, the person said.
