Google told employees it believed its work with the Department of Defense was consistent with its AI principles and said it was “taking an even greater focus” on national security work with the government, Business Insider found.
At a Google DeepMind town hall in January, leaders including CEO Demis Hassabis answered questions from staff about how Google makes sure its partnerships with the Department of Defense and Boston Dynamics don’t violate the company’s AI policies.
Tom Lue, Google DeepMind’s global vice president, said the company has a “robust process” in place to consider details about intended use cases and whether partners are protecting safety, liability, and privacy, according to a recording of the meeting reviewed by Business Insider.
Lu reminded staff that the company updated its AI principles in 2025. The change removed previous commitments not to use Google’s technology for weapons development or surveillance purposes.
“The north star of the analysis is whether the benefits significantly outweigh the risks,” Lu said.
The technology industry’s role in warfare and surveillance has been a hot topic in recent weeks following Anthropic’s feud with the Department of Defense. Anthropic drew red lines about how its AI could be used for warfare and surveillance, leading the Department of Defense to designate the startup as a “supply chain risk.” Anthropic subsequently filed suit against the Department of Defense over the de facto blacklisting.
At the same time, tech companies like Google, Amazon, and Oracle, some of which once staunchly opposed involvement in the business of war, are increasingly competing for lucrative defense contracts.
At a January town hall before Anthropic’s battle with the Pentagon, Lu said employees should expect more of these types of deals.
“The other thing I’d like to say is that this is an area where we’re going to put more effort into it. We’re talking to governments about national security concerns,” Lu told employees.
He said the company is discussing cybersecurity and biosecurity risks.
Reached for comment, a Google DeepMind spokesperson pointed Business Insider to a blog post published last week. Learn more about Google’s latest Department of Defense contract. The tool is described as being used for tasks such as document creation and project planning.
Google rebuilds relationship with Department of Defense
Hassabis, who had previously worried that Google would use DeepMind’s technology for war, told the same City Hall officials that he was “very happy” with the balance Google was striking.
“It’s obviously a very complex world, as we all can see, but I think it’s our responsibility to work with democratically elected governments to provide world-class, unique capabilities to make the world safer and benefit the world,” he said.
After pulling out of the military pact in 2018 amid an employee revolt, Google re-engaged with the Pentagon last year and secured contracts for AI and cloud products. This month, it won a contract to deploy AI agents across the department’s unclassified network.
Company executives say these deals primarily involve using AI for administrative tasks and do not play a role in identifying, tracking or attacking targets. Lu said during the meeting that the Pentagon contract includes tasks such as summarizing information, extracting text from contracts and other “back-office-type tasks.”
Employees have expressed concerns that Google won’t have full control over how its partners end up using the technology, such as the AI tools it provides to the Israeli government. Last month, Google and OpenAI employees signed an open letter urging them not to allow their technology to be used for autonomous killing or mass surveillance.
The Washington Post previously reported that a whistleblower complaint filed with the SEC alleges that Google violated its own policies by assisting an Israeli drone contractor in analyzing video in 2024. A Google spokesperson told the Post that the use of the company’s AI services at that time was too low to be “meaningful.”
Government contracts aren’t the only concern for some Googlers. Google DeepMind announced in January that it was working with Boston Dynamics to integrate Gemini into one of its robots.
“Boston Dynamics has been very clear that these technologies cannot be used for weaponized purposes. We have very robust processes in place,” Lu said at City Hall.
Have something to share? To contact this reporter via email, please specify the following address: hlangley@businessinsider.com Or text us at 628-228-1836. Use a personal email address and non-work device. Here’s a guide to sharing your information securely.