Overview
- Multiple outlets reported Thursday that Google is negotiating a deal to run its Gemini models in classified Defense Department systems under an “all lawful purposes” framework with contract bans on domestic mass surveillance and weapons that act without human control.
- The talks include adding GPU racks to Google Distributed Cloud and enabling Google’s tensor processing units, its custom AI chips, to run inside accredited classified environments for the first time.
- Alphabet and the Pentagon have not confirmed the negotiations, and the effort reportedly builds on a November agreement that covered unclassified use of Google’s models for all lawful purposes.
- The push comes after the Pentagon’s dispute with Anthropic, which refused to loosen safety limits and was labeled a supply‑chain risk, as OpenAI secured a separate deal with similar guardrails that its CEO urged the Pentagon to apply to all vendors.
- Google’s reentry into defense work marks a shift from its 2018 Project Maven retreat, yet AWS and Microsoft still run most classified workloads, which could constrain how quickly Google can provide capacity.