On February 23, 2026, Axios reported, citing a defense official, that Elon Musk’s xAI had signed an agreement allowing the US military to use Grok inside classified systems. This is not routine procurement news. In the military’s most sensitive classified environments, Anthropic’s Claude had been the only authorized model; Grok is the second. And the clause that bought xAI its ticket — an “all lawful use” standard — is exactly the one Anthropic had refused to sign.
The story matters because it turned a model vendor’s safety red lines into the explicit price of market access. Whoever accepts the government’s use terms gets the contract; whoever refuses faces the threat of being cut off. Defense AI market share is now being decided by how much usage restriction each lab is willing to absorb.
The Deal: “All Lawful Use” Buys the Ticket
Per the original Axios report and follow-up coverage from Türkiye Today and Anadolu Agency on February 24, the core of the agreement is that xAI accepted the Pentagon’s “all lawful use” standard. Grok already lived in unclassified military systems alongside Google’s Gemini and OpenAI’s ChatGPT; the news here is crossing the classified line.
Claude had supported intelligence analysis, weapons development, and battlefield operations inside those most-sensitive environments, making the Pentagon’s AI capability partly a Claude capability. Grok becoming the first second model with access to that tier of defense infrastructure is more symbolic than operational for now — the agreement is signed, but the integration has not started.
Why Anthropic Is Standing Outside
The dispute started over usage terms. The Pentagon threatened to sever ties with Anthropic over Claude’s usage limits; Anthropic refused the “all lawful purposes” demand and held two red lines: no mass surveillance of Americans, and no fully autonomous weapons.
According to the February 24 reporting, Defense Secretary Pete Hegseth was expected to meet Anthropic CEO Dario Amodei on Tuesday with an ultimatum. The heavier lever is a “supply chain risk” designation — if Anthropic gets that label, defense contractors could be forced to cut all ties with the company, an impact that reaches far beyond the Pentagon’s own contracts.
How Hard Is It to Replace Claude?
Officials themselves admit the technical replacement is not simple. Claude is deeply integrated into Pentagon workflows, including work with Palantir. Whether Grok can fully replace Claude, and how long a transition would take, remained open questions in the reporting. The realistic near-term picture is a two-vendor setup: Claude stays inside existing integrations while Grok seeps in through new workloads.
The lesson for product teams is blunt: government-grade AI procurement is not a one-time model choice but a long commitment bound to integration depth. Ripping out a model already embedded in workflows costs about as much as rewriting them.
Google and OpenAI Are Still Queuing
The fuller picture from Axios and follow-ups: Google and OpenAI are actively chasing their own classified seats. The New York Times reported Google was close to a classified Gemini deal while OpenAI was not; a defense official pushed back, saying talks with both remain ongoing. The common denominator is the gate itself — anyone entering classified systems must first accept the “all lawful purposes” requirement.
This is the government-side version of the terms-tightening trend we flagged in our opening outlook for 2026: commercial API terms and government use terms are hardening at the same time, squeezing model vendors from both ends.
What It Means for Vendors and Procurement
Three observations. First, safety red lines now have an explicit price tag — Anthropic traded two usage limits for the risk of losing its exclusive classified seat, while xAI traded blanket acceptance for first-mover access; both paid, just in different currencies. Second, defense AI procurement is shifting from single-vendor to competitive multi-vendor sourcing, which hands contractors and systems integrators a new architecture problem: abstraction layers and swapability will become real bid requirements. Third, political risk is now a first-class variable in the AI supply chain — expect continued scrutiny of Musk’s overlapping business interests, and treat policy volatility as part of any vendor-risk assessment that includes xAI.
For working developers the signal is simpler: as capability gaps between frontier models narrow, procurement weight shifts to terms, trust, and integration depth. Choosing a model vendor increasingly looks like a policy statement.
Sources
- Musk’s xAI, Pentagon reach deal to use Grok in classified systems — Axios
- Pentagon signs deal with Musk’s xAI to use Grok in classified military systems — Türkiye Today
- Pentagon, Musk’s xAI reach agreement to use Grok in classified systems: Report — Anadolu Agency
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.