the government explicitly retained the right to override it
In a quiet but consequential shift, Google has joined the Pentagon's classified AI program, lending its most capable systems to a military apparatus now officially renamed the Department of War. The agreement, broad enough to cover weapons targeting and mission planning, places Silicon Valley's most powerful tools in the hands of those who wage conflict — even as the engineers who built those tools have warned they are not yet ready for such weight. It is a familiar human story: the instrument outpaces the wisdom required to wield it, and the contracts meant to hold the line are written by the same hands that open the gate.
- Google has quietly signed a classified deal granting the Pentagon access to its AI for weapons targeting and mission planning — a threshold once treated as a red line by the tech industry.
- The contract allows the military to adjust or disable Google's own safety filters on request, effectively handing the government a key to the guardrails Google built.
- Safeguard language prohibiting autonomous weapons without human oversight exists in the agreement, but the same document strips Google of any right to veto how the government actually deploys the technology.
- The Pentagon has now secured $200M+ AI agreements with Google, OpenAI, xAI, and Anthropic — assembling a classified arsenal of machine intelligence even as its creators warn the systems remain error-prone in high-stakes decisions.
- The deal moves AI from the lab into live military operations, leaving the adequacy of contractual safeguards as the last line of defense between oversight and autonomous lethality.
Google has entered a classified defense contract with the Pentagon, joining OpenAI and Elon Musk's xAI in a small but growing circle of AI firms supplying machine learning systems directly to the U.S. military. The agreement authorizes use of Google's AI for "any lawful government purpose" — language expansive enough to include weapons targeting and mission planning on classified networks.
The Pentagon signed deals worth up to $200 million each with major AI laboratories throughout 2025, reflecting a deliberate strategy to maximize flexibility in how artificial intelligence is deployed across defense operations. The Google contract goes a step further than most: it grants the military the right to request modifications to Google's built-in safety settings and content filters, giving the government authority to reshape the AI's behavior according to operational need rather than the parameters its creators designed.
The agreement does include language committing both parties to avoid autonomous weapons deployment without meaningful human oversight, and to refrain from domestic mass surveillance. But the same contract clarifies that this language "does not confer any right to control or veto lawful Government operational decision-making" — meaning the safeguards are aspirational rather than enforceable, and Google has no mechanism to prevent uses that contradict them.
A Google Public Sector spokesperson described the deal as an amendment to an existing contract rather than a new arrangement. Neither Google nor the Pentagon — now officially renamed the Department of War under the Trump administration — offered further comment. The episode crystallizes a deepening tension: the military is racing to operationalize AI while the technology's own architects continue to warn it is not yet reliable enough for decisions that carry lethal consequence.
Google has become the latest major artificial intelligence company to enter into a classified defense contract with the Pentagon, according to reporting from The Information on Tuesday. The agreement grants the U.S. military access to Google's AI systems for what the contract describes as "any lawful government purpose"—a formulation broad enough to encompass weapons targeting, mission planning, and other sensitive military operations conducted on classified networks.
The deal places Google alongside OpenAI and Elon Musk's xAI in a small but growing roster of AI firms now supplying machine learning models directly to the Department of Defense for classified work. The Pentagon signed agreements valued at up to $200 million each with major AI laboratories during 2025, including Anthropic, OpenAI, and Google. These contracts reflect a deliberate Pentagon strategy to maintain maximum flexibility in how it deploys artificial intelligence across defense operations, even as the technology's own creators have cautioned against relying on AI systems that remain prone to error when making decisions about weapons deployment.
The Google contract includes a notable provision: the Pentagon retains the right to request adjustments to the company's AI safety settings and content filters. This means the military can ask Google to modify or disable safeguards that the company has built into its systems. The arrangement essentially gives the government authority to reshape the AI's behavior according to its operational needs, rather than being constrained by the safety parameters Google designed.
The agreement does contain language acknowledging certain boundaries. Both parties agreed that the AI system should not be used for domestic mass surveillance or for autonomous weapons—including automated target selection—without meaningful human involvement and control. Yet the contract simultaneously stipulates that this language "does not confer any right to control or veto lawful Government operational decision-making." In other words, the safeguard exists on paper, but the government retains final authority over how the system is actually used, and Google has no mechanism to prevent deployment that contradicts the stated restrictions.
A spokesperson for Google Public Sector, the division that manages the company's relationships with U.S. government agencies, characterized the new agreement as an amendment to an existing contract rather than an entirely new arrangement. Neither Google nor the Pentagon—now officially renamed the Department of War under the Trump administration—provided immediate comment when contacted by news organizations.
The deal underscores a fundamental tension in the current moment: the Pentagon is racing to integrate advanced AI into military operations while technology companies have publicly warned that their systems are not yet reliable enough for high-stakes applications like weapons targeting. The Pentagon's strategy appears to be to secure access to the most capable AI available and then manage the risks through contractual language and human oversight protocols. Whether those safeguards will prove adequate in practice remains an open question as these systems move from testing into operational use.
Notable Quotes
The Pentagon is seeking to preserve all flexibility in defense and not be limited by warnings from the technology's creators against powering weapons with unreliable AI— The Information reporting
The Hearth Conversation Another angle on the story
Why does it matter that Google specifically signed this deal now, when OpenAI and xAI already have similar contracts?
Because Google is the largest of the three by market capitalization and technical reach. It signals that the Pentagon isn't relying on any single AI vendor—it's building redundancy and competition into its defense infrastructure. That's both reassuring and concerning.
Reassuring how?
If one company refuses a request or experiences a breach, the Pentagon has alternatives. Concerning because it means the military is committed to this path regardless of any single company's hesitation.
The contract says the AI shouldn't be used for autonomous weapons without human control. How is that different from what they're already doing?
It's not really different. The language is there, but the government explicitly retained the right to override it. The safeguard is more symbolic than binding.
So why include it at all?
Liability, probably. And political cover. If something goes wrong, Google and the Pentagon can both point to the contract and say they agreed on principles. But principles without enforcement are just words.
What happens if the AI makes a targeting error?
That's the question nobody wants to answer yet. The contract doesn't specify who bears responsibility—Google, the Pentagon, or both. That ambiguity will likely be tested in court eventually.
And the safety filters Google can be asked to disable?
That's the real leverage point. It means the Pentagon can essentially ask Google to remove the guardrails the company built in. Google agreed to do it. That's the story.