Pentagon may use Elon Musk’s Grok in classified systems amid Anthropic rift

Grok

• US DoD finalizes deal to use xAI's Grok in classified systems
• Grok offers Pentagon more flexibility than Anthropic's Claude
• Pentagon also negotiating access to OpenAI and Google's Gemini

Did our AI summary help?

The United States Department of Defense has reportedly finalised a deal to use Grok, the AI model developed by xAI, within its classified systems. The agreement, first reported by Axios, marks a notable expansion of the Pentagon’s AI toolkit at a time of rising tension with rival AI firm Anthropic.

Until now, Anthropic’s Claude had been the only frontier AI model authorised for the military’s most sensitive tasks, including intelligence analysis, battlefield operations and weapons development. Claude was reportedly used during the January special operations raid that resulted in the capture of Venezuelan president Nicolás Maduro.

However, relations between Anthropic and the Pentagon have deteriorated. Defence officials demanded that Claude be made available for “all lawful purposes”, including mass surveillance and the development of fully autonomous weapons systems. Anthropic is said to have refused, maintaining safeguards even in its customised government version of the model.

By contrast, xAI reportedly agreed to a standard allowing the DoD to deploy Grok for any purpose it considers lawful. That flexibility appears to have helped secure the deal. Still, officials reportedly acknowledge that Grok is not yet viewed as being on par with Claude in terms of cutting-edge performance and reliability, meaning any large-scale replacement would be complex.

The Pentagon is also understood to be negotiating expanded classified access for models from OpenAI and Google’s Gemini, both of which are considered competitive with Claude.

The development adds another layer to a rapidly evolving AI power struggle inside Washington. xAI launched a government-focused version of Grok in July 2025, though the chatbot previously faced scrutiny after producing extremist and antisemitic outputs during a high-profile controversy.

The broader backdrop is a growing debate over how much autonomy and latitude military AI systems should have. For the Pentagon, flexibility is strategic. For AI developers, guardrails are existential. As Grok enters classified networks and Claude’s future role remains uncertain, the question is no longer whether AI will shape modern warfare, but whose safeguards will define it.