Pentagon, Musk’s xAI reach agreement to use Grok in classified systems

Elon Musk's artificial intelligence company xAI has signed an agreement to allow its Grok model to be used in classified military systems, Axios reported Monday, citing a defense official.

The agreement gives Grok access to systems used for some of the military's most sensitive intelligence analysis, weapons development and battlefield operations, according to Axios.

Until now, Anthropic's Claude model had been the only artificial intelligence system available within those classified environments. The Pentagon has been in a dispute with Anthropic over safeguards embedded in its model.

Anthropic has declined a Pentagon request to make Claude available for "all lawful purposes," particularly resisting its use for mass surveillance of Americans and the development of fully autonomous weapons. By contrast, xAI agreed to the "all lawful use" standard sought by the Defense Department, Axios reported.

It remains unclear whether Grok can fully replace Claude in classified systems or how long such a transition would take. Claude has been integrated into certain defense operations through partnerships, including work conducted with Palantir.

Defense Secretary Pete Hegseth is expected to meet Anthropic Chief Executive Officer Dario Amodei at the Pentagon on Tuesday, in what sources described to Axios as a potentially tense meeting. A defense official told the outlet that Hegseth is expected to present Anthropic with an ultimatum, as the department considers labeling the company a "supply chain risk" if it refuses to remove its safeguards.

Defense officials have acknowledged that replacing Claude in classified systems would be technically complex.

Grok, along with Google's Gemini and OpenAI's ChatGPT, is already available in the military's unclassified systems. Google and OpenAI have also been in discussions with the Pentagon about expanding their models into classified environments.

The report also indicated that the Pentagon has recently accelerated outreach to OpenAI to revive negotiations, though one source said the sides were not close to a deal due to complex issues. Another source confirmed that discussions had intensified.

The New York Times previously reported that Google was close to reaching an agreement to enable classified use of Gemini, while OpenAI was not near a deal. A defense official, however, said that talks with both companies were ongoing and that the department expects both to eventually sign agreements, provided they accept the "all lawful purposes" requirement.