OpenAI amends Pentagon deal to ban domestic mass surveillance

It took less than a week for OpenAI's defense deal to turn into a full-blown PR nightmare.

After facing employee pushback, protests outside its San Francisco headquarters, and a wave of user cancellations, OpenAI CEO Sam Altman announced that the company will amend its agreement with the Department of Defense. The new terms aim to explicitly prohibit using its AI for domestic mass surveillance against Americans.

Altman published an internal memo on X, admitting the company rushed the contract on Friday, February 27, and conceded that the move looked opportunistic and sloppy.

Walking back the Pentagon deal

The amended terms cite the Fourth Amendment, the National Security Act of 1947, and the Foreign Intelligence Surveillance Act of 1978. They state that OpenAI systems cannot be intentionally used for domestic surveillance of U.S. persons, including through purchasing commercial personal data.

Altman claimed that intelligence agencies like the NSA cannot use OpenAI services without further contract modifications. He added that if handed an unconstitutional order, he would rather go to jail than obey it.

OpenAI research scientist Noam Brown clarified that the company will not deploy to the NSA or other defense intelligence agencies for now while legal loopholes are addressed.

This marks a sharp pivot from days earlier, when OpenAI jumped in within 24 hours after President Trump ordered all government agencies to stop using Claude, following Anthropic's refusal to drop guardrails against mass surveillance and autonomous weapons.

A massive consumer backlash

Critics immediately questioned whether the original terms offered any real protection. Sources familiar with negotiations told The Verge that OpenAI's agreement hinged on complying with "any lawful use," a standard intelligence agencies have historically stretched to cover sweeping domestic programs like PRISM.

OpenAI's former policy research head, Miles Brundage, openly criticized the move, suggesting staff should assume OpenAI caved while framing it as holding the line.

The consumer fallout was swift. Sensor Tower data showed ChatGPT uninstalls jumped 295 percent day-over-day.

Meanwhile, Anthropic shot to the number one free app on the App Store, leaping past both ChatGPT and Google Gemini. Anthropic quickly rolled out a memory import tool to help defecting users bring their chat history along.

Why it matters

For regular users, this standoff makes one thing clear: enterprise AI contracts directly impact consumer trust. If you rely on ChatGPT for sensitive personal or business tasks, watching how loosely privacy terms get drafted during government negotiations is a valid reason to pause.

For the AI industry, this breaks the illusion that frontier labs can quietly partner with defense departments without alienating their mainstream consumer base. OpenAI tried to play realpolitik by stepping into the void Anthropic left, but it misjudged public perception and its own workforce.

We think OpenAI's hasty backtrack exposes a severe strategic miscalculation. Anthropic took the financial hit and government wrath up front, winning immense brand loyalty and product momentum in return. OpenAI grabbed the contract, but it traded away years of goodwill for an agreement it is now scrambling to rewrite in public.