Google signs classified AI contract with Pentagon

The Information reported, citing a person familiar with the matter, that Google has signed an agreement with the U.S. Department of Defense to allow use of Google's AI models for classified work. Reuters carried that reporting and noted the agreement, according to The Information, allows the Pentagon to use Google's AI for "any lawful government purpose" and requires Google to help adjust the company's AI safety settings and filters at the government's request. Reuters further reported that the contract contains language stating the AI system is not intended for domestic mass surveillance or autonomous weapons (including target selection) without appropriate human oversight, while also adding that "the Agreement does not confer any right to control or veto lawful Government operational decision-making." Reuters said it could not verify the report independently. The Information reported that a Google Public Sector spokesperson described the new agreement as an amendment to its existing contract. Reuters also referenced prior Pentagon agreements worth up to $200 million each in 2025 with major AI labs, including Anthropic, and noted that OpenAI and xAI have arrangements to supply models for classified use.

Companies providing models for classified networks are typically required to support integration with isolated or accredited environments and to implement configuration controls that align with government security baselines. Industry reporting around similar Pentagon arrangements highlights three recurring operational requirements:

• •vetted access into classified networks,
• •configurable safety and filtering controls that can be adjusted per customer rules,
• •contractual clauses limiting certain uses while preserving operational authority for the government.

These are industry patterns observed in public reporting on previous contracts; they are not claims about Google's internal implementation beyond the cited coverage.

Reports that major labs including Google, Anthropic, OpenAI, and xAI are providing models for classified work underscore expanding defense-sector access to advanced models. This dynamic raises trade-offs between operational flexibility for defense customers and externally stated safety guardrails from model providers, a tension visible in the contract language Reuters quoted. The Reuters account also notes a high-profile political context, stating President Donald Trump has renamed the Pentagon the Department of War, which reporters mentioned while covering the story.

For practitioners and observers: monitor formal contract filings or public statements from Alphabet/Google and the Department of Defense for confirmation and contract text; watch for specifics on deployment environments (accredited classified networks versus stovepiped enclaves), auditing and logging requirements, and any technical annexes that define allowed and disallowed model behaviors. Also track whether future reporting reveals how safety-setting adjustments will be governed and audited.