After Iran strikes, Google, OpenAI employees demand limits on military AI contracts

Artificial Intelligence

• Tech workers call for limits on military AI after US-Iran strikes
• Nearly 900 sign open letter, most from Google and OpenAI
• Google faces renewed internal dissent over Pentagon AI talks

Did our AI summary help?

Tech workers at Google, OpenAI and other major firms are escalating internal pressure over military AI contracts, circulating open letters that call for firm boundaries on how their employers engage with the Pentagon.

The backlash intensified after U.S. strikes on Iran over the weekend and the Department of Defense’s decision to designate Anthropic a “supply chain risk,” effectively blacklisting its models from federal use. Anthropic had reportedly refused to permit its technology to be deployed for mass surveillance or fully autonomous weapons.

One open letter, titled “We Will Not Be Divided,” grew from a few hundred signatures on Friday to nearly 900 by Monday. Close to 800 signatories are from Google and almost 100 from OpenAI. The letter accuses the Defense Department of attempting to pressure AI companies individually, warning that division among firms would weaken their ability to push back.

“They’re trying to divide each company with fear that the other will give in,” the letter states. “That strategy only works if none of us know where the others stand.”

Google in the spotlight again

For Google, the moment feels familiar. The company is reportedly in talks with the Pentagon to bring its Gemini AI model onto classified systems — a move that has revived memories of its 2018 internal revolt over Project Maven, a drone-analysis programme that ultimately collapsed after employee protests.

More than 100 Google employees working on AI reportedly signed a separate internal letter last week urging leadership to draw the same red lines as Anthropic. According to reports, they specifically asked the company to reject uses tied to mass surveillance or autonomous weapons.

Jeff Dean, Google’s chief scientist, appeared sympathetic to at least part of the criticism. In a public thread, he wrote that mass surveillance violates the Fourth Amendment and carries a chilling effect on freedom of expression, adding that such systems are prone to political or discriminatory misuse.

The tension builds on previous flashpoints. In 2024, Google dismissed more than 50 employees after protests over Project Nimbus, a $1.2 billion cloud contract with the Israeli government. Executives maintained the deal complied with the company’s AI Principles, though reports indicated the agreement allowed AI tools including image categorisation and object tracking for state entities.

Early last year, Google reportedly revised its AI Principles and removed language explicitly prohibiting “building weapons” or “surveillance technology,” a shift that has fuelled renewed scepticism among staff.

Wider industry pushback

Beyond Google and OpenAI, hundreds of workers across the industry — including employees affiliated with Salesforce, Databricks, IBM and Cursor — signed another letter urging the Department of Defense to withdraw its designation of Anthropic as a supply chain risk.

That letter calls on Congress to examine whether such extraordinary authorities should be used against an American technology company that declined certain military terms. It argues private firms should not face retaliation for refusing government demands they view as ethically problematic.

Activist coalition No Tech For Apartheid also weighed in, urging cloud leaders — including Google, Amazon and Microsoft — to reject Pentagon contract terms that could enable mass surveillance or abusive AI deployments. The group specifically cited the possibility of Gemini being deployed in classified environments, drawing comparisons to Elon Musk’s xAI securing defence access for its Grok model.

While Anthropic and OpenAI have made public statements clarifying the scope and safeguards of their Pentagon negotiations, Google’s parent company Alphabet has remained largely silent.