Google signs Pentagon AI deal
- Google is nearing a deal to let the Pentagon use its artificial intelligence models for classified work, as employees press Sundar Pichai to stop it. - An open letter signed by hundreds of workers warned Google's tools could enable lethal autonomous weapons or mass surveillance if deployed in secret settings. - The fight echoes Google's 2018 Project Maven revolt and a wider Pentagon push for military AI access. (cbsnews.com)
Hundreds of Google employees are urging Chief Executive Sundar Pichai to block a Pentagon deal that would let the military use Google's artificial intelligence in classified settings. (cbsnews.com) The workers said in an open letter on April 27 that Google's AI systems are not appropriate for "classified workloads" and could be used in lethal autonomous weapons or mass surveillance. Google and the Pentagon did not immediately respond to CBS News requests for comment on the letter. (cbsnews.com) CBS News said the letter followed reporting from The Information that Google is negotiating with the U.S. Department of Defense over a possible classified AI arrangement. The reported talks would expand Google's role beyond public or lower-sensitivity government work into secret military environments. (cbsnews.com) (theinformation.com) Classified AI work means a model is made available on secure government networks used for sensitive military and intelligence tasks. In practice, the Pentagon has been using large language models to summarize documents, analyze video and images, and speed up battlefield planning. (cbsnews.com) Former Pentagon officials and military analysts told CBS News that artificial intelligence can sift through more data than human teams can process on operational timelines. Retired Admiral Mark Montgomery said the military is now processing roughly 1,000 potential targets a day in some operations, with humans still in the loop. (cbsnews.com) The dispute at Google lands as the Pentagon is pushing harder on access terms with AI companies. CBS News reported in February that Defense Secretary Pete Hegseth demanded unrestricted military access to Anthropic's Claude model and that officials discussed using the Defense Production Act. (cbsnews.com) That fight ended with Anthropic being pushed out of Pentagon operations after seeking written guardrails against mass surveillance of Americans and fully autonomous targeting decisions. CBS News reported that Google, OpenAI and xAI had each received Pentagon contracts worth $200 million last year. (cbsnews.com) Google's own published AI principles, first released in 2018, say the company will not pursue technologies whose principal purpose is to cause injury or surveillance that violates internationally accepted norms. Google said in its 2026 Responsible AI Progress Report that it is still applying those principles to product and research decisions. (blog.google 1) (blog.google 2) The employee backlash also revives a long-running fault line inside Google over military work. In 2018, worker protests over Project Maven, a Pentagon drone-imagery program, helped force the company to step back and publish those AI principles. (blog.google 1) (blog.google 2) What happens next is whether Google signs the deal anyway and, if it does, what limits are attached to the model's use on classified networks. The employee letter asks Pichai to answer that question before the Pentagon does. (cbsnews.com)
