Google Signs A.I. Deal With the Pentagon | UNW

Google said on Tuesday that it had signed a deal to provide the Pentagon with its artificial intelligence models for classified work, amid a dispute between the Department of Defense and the A.I. start-up Anthropic over how to responsibly use the technology during war.

The agreement is part of a $200 million contract that the search giant signed last year with the Pentagon to provide it with A.I. tools. Now the Pentagon can use Google’s A.I. on classified systems for “any lawful governmental purpose,” people with knowledge of the deal said. The language mirrored that of deals the Department of Defense struck last month with OpenAI and Elon Musk’s xAI to use their A.I. models on classified networks, people who had reviewed the contracts said.

“We are proud to be part of a broad consortium of leading A.I. labs and technology and cloud companies providing A.I. services and infrastructure in support of national security,” said Jenn Crider, a Google spokeswoman. “We remain committed to the private and public sector consensus that A.I. should not be used for domestic mass surveillance or autonomous weaponry without appropriate human oversight.”

U.S. officials confirmed that Google and the Pentagon had reached an agreement that would allow the company’s Gemini A.I. model to be used on a wide array of classified systems. The agreement — along with those with OpenAI and xAI — could give the military flexibility and avoid a situation where any single company has a lock on contracts, one of the officials said.

Google declined to comment on contractual details. A Pentagon official confirmed that the contract with Google had been signed but declined to provide details. The Information earlier reported details of the deal.

The Pentagon has moved aggressively to secure agreements with Silicon Valley’s biggest A.I. companies since January, when Defense Secretary Pete Hegseth announced that the technology should be widely integrated across the military.

The move has led to friction between the Pentagon and some Silicon Valley companies over the role of technology in war. Anthropic, the start-up whose A.I. technology was first used on classified networks, refused to remove ​guardrails against using its A.I. for autonomous weapons or domestic surveillance. The Pentagon countered that a company should not be telling it how to use A.I.

In March, the Pentagon designated Anthropic a “supply chain risk,” a label that effectively barred the San Francisco company from working with the federal government. Anthropic has since sued the Pentagon over the designation, and is seeking to continue working with other parts of the federal government.

White House officials met with Anthropic about 10 days ago to discuss a compromise on how the government could use the company’s A.I. This month, Anthropic released a powerful A.I. model called Mythos, which many believe could be critical for security.

Defense officials said A.I. was becoming increasingly central to their work. Last week, the Pentagon asked Congress for $2.3 billion to expand Project Maven, a system built by Palantir, the data analytics company, to integrate A.I. technology into the military. The system uses Anthropic’s A.I. technology to help analyze intelligence. If Anthropic and the Pentagon are unable to compromise, another company’s A.I. technology could be introduced instead, two officials familiar with the system said.

Google’s Gemini is already in use in unclassified systems at the Department of Defense. Bringing in new A.I. models has created some technical hurdles, as government cloud providers have had to change configurations or update hardware, officials have said.

Google’s deal with the Pentagon is the latest validation of the Silicon Valley company’s A.I. technology. After slipping behind upstarts like OpenAI and Anthropic, the search giant — which was a pioneer in artificial intelligence — caught up in performance and features with the latest Gemini release.

Google enjoys advantages its rivals don’t have. It has deep pockets, designs its own A.I. chips and has a cloud computing business with data centers spread across the world.

Google struck the deal with the Pentagon despite opposition from some employees. On Monday, more than 600 from the company’s A.I. and cloud divisions signed a letter urging Google to deny the Pentagon use of its technology for classified military operations. More than 100 working on artificial intelligence had also signed a letter in February calling on Google to establish the same safeguards in its government contracts that Anthropic was seeking.

In 2018, Google struck an agreement with the Pentagon that caused an employee uprising and led the company to discontinue its contract. In the years that followed, the company centralized the decision-making process around those agreements and sought to quell employee activism.

Google and Anthropic are rivals but also partners. Google began backing Anthropic in its infancy. Last week, Google said it had committed to investing as much as $40 billion in Anthropic. Anthropic relies on Google and Amazon for the cloud computing it needs to develop A.I., and has become one of the largest users of the tech giants’ custom A.I. chips.

Kate Conger contributed reporting.