On October 20, 2026, the BBC reported that several Google DeepMind researchers had resigned over the company's military AI contracts. Sofia Liguori, a London-based research engineer, left after three years and said publicly, despite contract restrictions: "Realizing my company is pushing more and more for AI systems in military operations makes me feel ashamed to work at Google." Andreas Kirsch, an AI researcher, left earlier the same month, warning that current AI systems are not reliable enough for battlefield use and that Google "does not have sufficient control over how the Pentagon uses the technology, nor does it have full insight into its application."
A Google spokesperson said the company follows consistent review processes for these contracts, including compliance with its acceptable use policy and AI principles.
The problem is that those principles no longer say what the spokesperson implies.
The principle they cited was already deleted
Google published its AI Principles in 2018, after more than 4,000 employees signed a petition demanding a clear policy that neither Google nor its contractors would ever build warfare technology. About a dozen employees resigned in protest. Google dropped a Pentagon cloud bid, saying it could not be assured the contract would align with its principles.
The 2018 principles listed applications Google would not pursue. One category was "weapons or other technologies whose principal purpose or implementation is to cause or directly facilitate injury to people." Another was technology that gathers or uses information for surveillance violating internationally accepted norms.
In February 2025, that language disappeared from Google's updated AI principles page. CNN confirmed the removal via the Internet Archive. A blog post by James Manyika and Demis Hassabis justified the change by citing a "global competition for AI leadership within an increasingly complex geopolitical landscape" and the belief that "democracies should lead in AI development." The post did not mention weapons or surveillance.
Liguori's resignation letter invokes human oversight, international law and human rights. Kirsch says Google lacks control over Pentagon use. They are not describing a company that violated its principles. They are describing a company that removed them, while its public statements still gesture at the framework that once contained them.
600 signatures, then silence
The resignations are not the beginning of the story. They are the end of an internal process that failed.
On April 27, 2026, more than 600 Google employees signed an open letter to CEO Sundar Pichai, asking him to refuse any agreement allowing the Pentagon to use Google's AI for classified workloads. The letter included directors and vice presidents among its signatories. Its core argument was technical: "At this stage, there is no way to guarantee that our tools will not be used to cause terrible harm or violate individual freedoms, far from public scrutiny." The letter added: "The only way to guarantee that Google is not associated with such harm is to refuse any classified workload. Otherwise, such use could occur without our knowledge or power to stop it."
Google did not refuse. The contract proceeded.
Liguori was a signatory. In April she was still trying to change the company from within. In October she left. A Google spokesperson said there are dedicated channels for employees to raise concerns safely. The channels exist. Six hundred people used them. Google listened, then signed.
Anthropic kept the red line and lost the contract
The contrast that makes this story structural rather than moral comes from a different company.
In February 2026, the Pentagon gave Anthropic an ultimatum: remove restrictions on using Claude for fully autonomous lethal weapons and mass domestic surveillance, or lose the contract. Anthropic refused. Defense Secretary Pete Hegseth's department designated Anthropic a "supply chain risk" in March, canceling its military contracts and barring other Pentagon contractors from using its technology.
A federal judge in California later ruled the broader sanctions unlawful, finding the government had punished Anthropic for criticizing the Pentagon. But a separate appeals court in Washington upheld the supply chain risk designation, and that decision remains in effect.
The Pentagon turned to competitors. OpenAI signed an agreement the same day as Anthropic's ban. xAI signed in February. Google and Microsoft signed a few months later.
The incentive structure is visible in the sequence. Anthropic kept its red lines and was excluded. Google removed its weapons ban in February 2025, faced an internal letter in April 2026, and signed the contract anyway. Kirsch's claim that Google "has no right to control or veto how the Pentagon uses its AI" is not a failure of oversight. It is the terms of the deal.
P.S. The 2018 petition asked Google to never build warfare technology. Eight years later, the company's AI principles no longer contain the word "weapons," and the researchers who cited the old language are the ones leaving.
Frequently Asked Questions
Q: What exactly did Google remove from its AI principles?
A: In February 2025, Google removed language from its AI principles that promised not to pursue "weapons or other technologies whose principal purpose or implementation is to cause or directly facilitate injury to people" and surveillance violating international norms. CNN confirmed the removal via the Internet Archive.
Q: Who resigned and why?
A: Sofia Liguori, a London-based DeepMind research engineer, left after three years, saying she felt "ashamed to work at Google" as the company pushed AI into military operations. Andreas Kirsch, an AI researcher, left earlier, warning that Google lacks sufficient control over Pentagon use of its technology.
Q: What did the April 2026 employee letter ask for?
A: More than 600 employees, including directors and vice presidents, signed an open letter asking CEO Sundar Pichai to refuse any agreement allowing the Pentagon to use Google AI for classified workloads. They argued there is no way to guarantee the tools would not be used for harm far from public scrutiny.
Q: How did Google respond to the letter?
A: Google did not refuse the classified work. The company said it follows consistent review processes and complies with its acceptable use policy and AI principles, and that dedicated channels exist for employees to raise concerns.
Q: What happened to Anthropic in the same period?
A: Anthropic refused Pentagon demands to remove restrictions on autonomous lethal weapons and mass domestic surveillance. The Pentagon designated it a supply chain risk, canceling its contracts and barring other contractors from using its technology. A court later upheld that designation.
Q: Why does the contrast matter?
A: It shows the incentive structure. Anthropic kept its red lines and was excluded from military contracts. Google removed its weapons ban and signed. The resignations are not a story about individual conscience alone; they are a story about what the procurement system rewards.
