Pentagon and Commerce Dept. Play Hot Potato With AI Security Risk

On Thursday, Pentagon official Emil Michael issued a statement that should have sent Anthropic's investor relations team into convulsions: the company remains classified as a "supply chain risk" and therefore ineligible for federal contracts or sensitive partnerships. This declaration arrived approximately 24 hours after Commerce Secretary Howard Lutnick announced, with apparent confidence, that Anthropic was now "back on the right side." For a company whose valuation and commercial trajectory depend substantially on access to government dollars and institutional trust, this is not a rounding error in messaging. This is the regulatory equivalent of your bank calling to confirm your credit is excellent, then immediately freezing your account.

Anthropic, for those still keeping score at home, is one of the world's leading artificial intelligence laboratories, founded by former OpenAI leadership and backed by Google, Salesforce, and other heavyweight venture capitalists. The company builds large language models and AI safety research. It does not, to the public record, manufacture semiconductors, operate critical infrastructure, or engage in espionage. Yet within the Trump administration—which has made AI dominance a stated national security priority—there is apparently genuine disagreement about whether Anthropic represents a threat or an asset to U.S. interests. The optics are almost too perfect: a leading American AI company caught between two cabinet officials playing bureaucratic ping-pong with its supply chain status.

This is not Emil Michael's first rodeo with controversial government positions. His history of inflammatory statements and shifting alliances is well-documented. Yet his Thursday reaffirmation of Anthropic's blacklist status, mere hours after Lutnick's rehabilitative comments, suggests that Michael either did not receive the memo about administration alignment, or deliberately chose to ignore it. The latter interpretation is far more interesting, and far more damaging to Anthropic. It implies that even within the executive branch, the company is sufficiently controversial that different factions are willing to publicly contradict one another rather than present a unified front.

The stated reason: "supply chain risk." This phrase has become regulatory Vaseline—slippery, all-purpose, impossible to remove once applied. It is vague enough to mean almost anything (foreign investment exposure? algorithmic bias? the mere fact that Claude runs on Nvidia chips?) and specific enough to sound technical and therefore unimpeachable. Lutnick's counter-argument—that Anthropic is now "back on the right side"—is equally content-free. What side? Right of what? The subtext is that optics and political pressure, not technical security review, are driving these determinations.

For Anthropic's current and prospective customers in government and defense, this chaos is disastrous. No CISO is going to green-light a contract with a vendor whose regulatory status inverts every 24 hours. Federal procurement moves slowly for many reasons, but uncertainty about a vendor's legal standing is not one any agency will tolerate. Even if the Pentagon's blacklist is eventually reversed, the reputational damage is already priced in. The company has now been publicly associated with security concerns by one of the world's most powerful defense institutions, and that association will linger regardless of what Lutnick says next week.

What this story really exposes is the current state of AI regulation: not policy, but theater. Two cabinet officials contradict each other in real-time not because they operate from different technical briefs, but because the Trump administration has not actually decided what its AI policy is. Anthropic is simply the day's proxy for a much larger internal struggle about how America should relate to its own artificial intelligence companies. The company's investors are betting on an administration that can't even get dressed without contradicting itself. That's not valuation risk. That's a civilization risk.

In the startup world, we call this a Series Z: standing in front of two different judges who are both holding the hammer, trying to convince them you're a nail worth driving.