The Pentagon added ChatGPT and Grok to its military AI platform. Its own deadline to purge Claude -- the model a judge just ruled was illegally blacklisted -- lands the same week

September 1 is the date an Air Force Research Laboratory memo set for its own contractors to finish erasing Anthropic's Claude from their systems entirely -- a full month ahead of the Pentagon-wide deadline, and five days after a federal judge ruled the underlying blacklist illegal. Three days before that, on August 31, the Defense Department did the opposite: it activated OpenAI's ChatGPT Mil and xAI's Grok for Government on GenAI.mil, its in-house AI marketplace, growing the roster to three vendors serving roughly 1.7 million of the 3 million military and civilian personnel eligible to use it. The company notably absent from that roster is the same one whose court win the department is not treating as an instruction to stop.

The dispute traces back to a July 2025 contract that made Claude the first frontier model approved for use on classified U.S. government networks. When the Pentagon later asked Anthropic to waive the guardrails it had built into that deal -- restrictions against using Claude for mass domestic surveillance or fully autonomous weapons -- the company refused. Defense Secretary Pete Hegseth designated Anthropic a 'supply chain risk to national security' on February 27, 2026, a label ordinarily reserved for foreign adversaries and compromised hardware. U.S. District Judge Rita Lin ruled on August 27 that the designation was illegal, writing that the government's own filings amounted to punishing a critic rather than making a security case. The government is expected to appeal, and neither that ruling nor a separate, narrower case still pending in the D.C. Circuit has changed anything on the ground.

The purge mechanics were set in motion long before Judge Lin ever ruled. The Pentagon's chief information officer gave every branch and agency 180 days from April 2 -- until September 29 -- to remove Anthropic software from its systems. The Air Force Research Laboratory told its own contractors to move faster: identify every Anthropic product in their pipeline by August 1, and eliminate it entirely by September 1, reporting compliance through a Microsoft-hosted form. The memo said the earlier date exists 'to allow for administrative processing time and to ensure the deadline is met' -- a month of buffer against exactly the kind of legal reversal that arrived four weeks later.

GenAI.mil -- the secure portal that lets troops and civilians use commercial AI on controlled unclassified work without routing sensitive data through consumer apps -- launched with Google's Gemini for Government in December 2025. By August 31 it had grown to roughly 1.7 million unique users, and that day OpenAI's ChatGPT Mil and xAI's Grok for Government went live simultaneously, both cleared for Impact Level 5 handling (the accreditation tier for sensitive-but-unclassified data at enterprise scale). 'Access to a diverse suite of AI capabilities from across the resilient American technology stack will give warfighters the tools they need to act with confidence and safeguard the nation against any threat,' a War Department official said. ChatGPT Mil is built for document-heavy admin, logistics, planning and policy work; Grok for Government adds deep-reasoning inference, adaptive modes and reusable 'playbooks' meant to capture institutional knowledge. Claude wasn't invited to either announcement.

The reason isn't the litigation -- it's the same contract term that started it. Anthropic's objection was never to Pentagon business; the company has said its federal revenue already runs past $5 billion. What it refused to waive was the restriction against mass surveillance and autonomous-weapons use it built into its own July 2025 deal. The Pentagon wanted unrestricted access for 'lawful purposes' instead. (Judge Lin's ruling found the Pentagon's process unlawful; it never reached whether the underlying request itself was reasonable.) Until that gets resolved -- in a settlement, a new contract, or the pending D.C. Circuit case -- GenAI.mil's fourth vendor slot stays empty regardless of what happens to the blacklist itself.

What the ruling settled was narrow -- that the Pentagon's process was unlawful -- and what it left open is everything that actually determines whether Claude ever returns to a government desktop. Two systems are now running on separate clocks: a legal one, where Anthropic is winning, and an administrative one, where the purge keeps its own schedule regardless. Nothing announced this week suggests either side expects that to change soon.