The US Department of Defense is threatening to classify Anthropic as a supply chain risk. If that designation lands, every company that does business with the Pentagon — thousands of defence contractors — would be required to certify that they don’t use Claude in any of their workflows. That’s not a minor procurement dispute. It would be one of the most significant acts of AI supply chain control in US history.
Two Axios articles published on 15 and 16 February 2026 broke the full story. This is what’s happening, how it got here, and what it means for anyone using Claude — including via OpenClaw.
How It Escalated
Anthropic signed a contract with the Pentagon worth up to $200 million last summer. The deal was straightforward on paper: Anthropic would provide Claude for government and military use. The conflict emerged over the scope of “use.”
The Pentagon’s position, pushed by Pete Hegseth’s team, is that AI used in defence contexts must be available for “all lawful purposes.” That framing is doing a lot of work. In practice, it means the DoD wants Claude usable for surveillance, autonomous weapons systems, targeting assistance, and other defence applications without Anthropic imposing additional ethical constraints beyond what US law requires.
Anthropic’s position is that it maintains usage policies that restrict Claude from certain applications — specifically autonomous weapons and certain categories of surveillance — regardless of who the customer is. The company sees these restrictions as core to responsible AI deployment, not as optional features that can be contracted away.
That gap, between “all lawful purposes” and “lawful purposes minus our usage policy restrictions,” is where the relationship broke down.
The Venezuela Moment
The dispute became public — and emotional — when the Wall Street Journal reported that Claude had been used in the Maduro extraction operation in Venezuela. The operation, conducted through Anthropic’s partnership with Palantir, used AI assistance in planning. Anthropic was apparently not consulted in advance about this specific use.
When the Journal story landed, Anthropic reportedly sought clarification from the DoD about how their technology was being used. The Pentagon’s response, according to sources familiar with the matter cited by Axios, was essentially: this is what “all lawful purposes” means. If you don’t want us using it for operations like this, you shouldn’t be selling to us.
Anthropic pushed back. The company’s view is that it agreed to military use of Claude for legitimate administrative, logistics, and analysis purposes — not for use in active operations involving foreign nationals without prior notification.
The Venezuela incident crystallised what had been a simmering policy disagreement into an active standoff.
What “Supply Chain Risk” Actually Means
The threat on the table is the “Zero Anthropic” certification requirement for defence contractors.
Under this designation, companies that hold Pentagon contracts would be required to certify that Anthropic products — specifically Claude — are not used anywhere in their workflows. Not just on classified systems. Not just for defence-related work. Anywhere.
The scale of this is hard to overstate. Anthropic’s technology is embedded in hundreds of enterprise workflows, development environments, and productivity tools used by companies that also hold defence contracts. A certification requirement doesn’t just affect Anthropic directly — it forces an audit and potential rip-and-replace across an enormous number of secondary customers.
The supply chain risk designation is normally applied to foreign adversaries and companies with ties to hostile state actors. Applying it to a domestic AI company known for its safety-focused approach would be unprecedented. Which is precisely why the Pentagon is threatening it — it’s the nuclear option in contract negotiations, and using it would send an unmistakable signal to every AI company about the terms of doing business with the DoD.
Anthropic’s Dilemma
Anthropic is in a genuinely difficult position, and it’s worth understanding why this matters beyond corporate politics.
The company was founded on the premise that AI safety isn’t a marketing position — it’s a structural commitment baked into how Claude is trained and deployed. Dario Amodei and the safety-focused founders left OpenAI partly because they believed safety considerations were being overridden by commercial pressures. Accepting “all lawful purposes” as the standard for military use would represent exactly the kind of compromise they built Anthropic to avoid.
At the same time, the DoD contract was $200 million. And losing it — plus potentially being classified as a supply chain risk, chilling adoption across every other defence contractor — is an existential commercial threat for a company that has burned through enormous amounts of venture capital and hasn’t yet achieved profitability.
Bloomberg reported that talks about extending the contract are stalled. Sources described the atmosphere as tense. Nobody is saying it’s over, but nobody is saying it’s close to resolution either.
Why This Matters for OpenClaw Users
If you’re running OpenClaw and using Claude as your LLM backend, you’re not directly affected by a DoD procurement dispute. But there are second-order effects worth watching.
Model availability: If Anthropic loses the Pentagon contract and faces the supply chain designation, the company faces financial pressure that could affect its ability to maintain and develop Claude. Less competitive development means slower improvements to the model that powers many OpenClaw installations. Most OpenClaw users can [switch between models](/best-free-models-openclaw/) — OpenAI, Gemini, DeepSeek — but Claude’s strong reasoning and instruction-following is a genuine advantage for agent workflows.
Usage policy drift: If Anthropic is forced to relax its usage policies to retain government contracts, the guardrails that protect Claude from certain misuse categories could weaken. The current usage policy is one of Claude’s security features from an OpenClaw perspective — it limits what a compromised or prompt-injected agent can be instructed to do.
Geopolitical AI fragmentation: The dispute is a preview of what’s coming. As AI becomes infrastructure, governments will increasingly demand control over how commercial AI is used in sensitive contexts. That could mean usage reporting requirements, mandatory backdoors, or security certifications for AI providers. All of these downstream effects would touch OpenClaw users who rely on cloud-hosted models.
The Bigger Contest
The Pentagon-Anthropic dispute is happening in the context of a broader competition for AI talent and infrastructure. The same week this story broke, Peter Steinberger — OpenClaw’s creator — announced he was [joining OpenAI](/openclaw-creator-joins-openai/). Moonshot AI’s Kimi integration with OpenClaw went live. Manus Agents launched a no-setup alternative. The AI agent landscape is moving faster than procurement processes can handle.
The Pentagon’s position implicitly acknowledges something important: AI has become a genuine defence infrastructure question, not just a productivity tool. The fact that they’re threatening a domestic, safety-focused AI company with supply chain risk designation shows how seriously they take the question of who controls AI behaviour in operational contexts.
Anthropic’s position implicitly acknowledges something equally important: the line between “legitimate military use” and “using AI for things the company considers unethical” is real and worth defending, even at significant commercial cost.
Neither position is obviously wrong. But someone will blink first. The next few weeks will likely determine whether the US military’s AI stack is shaped by commercial providers maintaining independent ethical standards, or by providers who’ve accepted that the customer is always right.
Watch this space carefully. The outcome will set precedents that outlast any individual contract.
—
Sources: Axios (15 Feb 2026, 16 Feb 2026), Bloomberg (16 Feb 2026), TechCrunch (15 Feb 2026), The Hill (16 Feb 2026), Wall Street Journal (Venezuela operations report).