Google signed a confidential AI agreement with the Pentagon; employees issue an open letter opposing it

According to The Information, reported on April 28, Google has signed an agreement to provide the U.S. Department of Defense with artificial intelligence (AI) models for confidential work. The New York Times, citing people familiar with the matter, said the agreement allows the U.S. Department of Defense to use Google’s AI for lawful government purposes, with a nature similar to the classified AI deployment agreements the Pentagon signed with OpenAI and xAI last month.

Agreement Content and Broader Background

According to a report by The New York Times citing people familiar with the matter, the agreement between Google and the Department of Defense allows DoD to use Google’s AI models in confidential work. The Pentagon completed similar signings last month with OpenAI and xAI, allowing AI models to be deployed on classified networks; the Google agreement is the latest example of expanding AI cooperation for the Pentagon.

A Google spokesperson told The New York Times: “We are honored to be part of a broad coalition made up of leading AI labs, technology and cloud companies, delivering AI services and infrastructure for national security.” Decrypt asked Google for comment on this matter, and as of the time the report was published, Google had not responded.

Open Letter from Employees: Urging Refusal of Classified Workloads

According to a New York Times report, before the agreement was signed, hundreds of Google employees had already signed and sent an open letter to CEO Sundar Pichai, urging the company not to provide AI systems to the Pentagon for use in confidential work.

The open letter said: “We want AI to benefit humanity, not be used for inhumane or extremely harmful purposes. At present, the only way to ensure Google is not connected to such harms is to refuse any classified workloads. Otherwise, such uses may occur without our knowledge or ability to stop them.”

In the letter, “lethal autonomous weapons and mass surveillance” are listed as the main concerns, warning that “making the wrong decision now will cause irreversible damage to Google’s reputation, business, and standing in the world.”

Pentagon AI Integration Process and the Anthropic Case

According to a New York Times report, U.S. Defense Secretary Pete Hegseth said during a speech this January at Elon Musk’s Starbase that: “The United States must win the strategic competition for technological hegemony in the 21st century,” and added that “every non-classified and classified network in our department will soon deploy world-leading AI models.” Since then, the Pentagon has accelerated its pace of reaching agreements with major AI companies.

In related background, according to related reports, in March this year the Pentagon listed Anthropic as a “supply chain risk” company because Anthropic CEO Dario Amodei refused to allow the federal government unlimited use of its AI models; Anthropic then sued the Pentagon over that determination and sought to continue collaborating with other government departments.

Frequently Asked Questions

What kinds of uses does the classified AI agreement between Google and the Pentagon allow?

According to a report by The New York Times citing people familiar with the matter, the agreement allows DoD to use Google’s AI for “any lawful government purpose.” A Google spokesperson said the company insists that AI should not be used for domestic large-scale surveillance or autonomous weapons without human oversight.

Which AI companies has the Pentagon signed similar agreements with?

According to a report by The New York Times, last month the Pentagon signed agreements with OpenAI and xAI, respectively, allowing AI to be deployed on classified networks; Google’s agreement is the latest example of similar cooperation.

What is Anthropic’s position on similar issues?

According to related reports, in March this year the Pentagon listed Anthropic as a “supply chain risk” company because Anthropic refused to allow the federal government unlimited use of its AI models; Anthropic then sued the Pentagon over that determination and sought to continue collaborating with other government departments.