WASHINGTON — Just when it appeared that Anthropic had repaired its turbulent relationship with the Trump administration, the Pentagon has delivered a striking reminder that one of Washington’s most bitter artificial-intelligence battles is far from finished.
Emil Michael, the U.S. under secretary of defense for research and engineering, said Thursday that Anthropic remains designated as a “Supply Chain Risk” within the Defense Department and across the U.S. defense industrial base.
His statement directly complicated comments made only a day earlier by Commerce Secretary Howard Lutnick, who said Anthropic had returned to the administration’s good graces.
Speaking to Axios during a G20 technology gathering in North Carolina, Lutnick said Anthropic was “back on the right side” and added that the administration now trusts the company.
That leaves one of Silicon Valley’s most important AI developers facing an extraordinary contradiction: one senior Trump administration official says Washington trusts Anthropic, while another says the company behind Claude remains a risk to America’s defense supply chain.
And the disagreement comes less than a week after Anthropic won a major court battle against the Pentagon.
A Federal Judge Ruled Against the Pentagon
On August 27, U.S. District Judge Rita F. Lin sided with Anthropic in its lawsuit challenging the government’s actions against the company.
Lin ruled that the Pentagon’s measures amounted to unlawful retaliation in violation of the First Amendment and that Anthropic had been denied required due-process protections under the Fifth Amendment.
The court also determined that Defense Secretary Pete Hegseth’s decision to classify Anthropic as a supply-chain risk violated the governing statutory framework and was arbitrary and capricious.
The ruling went even further in examining the government’s justification.
Under the statute invoked by the Pentagon, a supply-chain risk concerns the danger that an adversary could sabotage, secretly introduce unwanted functions or otherwise compromise a protected system.
But Lin found that Anthropic’s public disagreement with the Pentagon over contractual restrictions did not fit that definition.
The court said the relevant law was aimed at risks such as covert sabotage or malicious code—not an American technology company publicly insisting on restrictions governing how its product could be used.
The Pentagon had argued that continuing dependence on Anthropic’s technology created potential risks because AI models can be modified or updated and because a supplier could theoretically influence their behavior.
Court documents show Michael had raised concerns including model “drift,” manipulation and the possibility of the Pentagon depending on what he characterized as a technological “black box.”
The judge, however, concluded that the administrative record did not establish the kind of supply-chain threat required by the law.
What Started the Anthropic-Pentagon Fight?
At the center of the dispute is a much larger question facing governments around the world:
Who ultimately gets to decide what an advanced AI system can be allowed to do—the company that created it or the military buying it?
Anthropic has worked with the U.S. government and has repeatedly said it wants Claude to support American national security.
But the company drew two major red lines.
Anthropic said it would not remove safeguards designed to prevent Claude from being used for certain forms of mass domestic surveillance or for fully autonomous lethal weapons without adequate safeguards.
CEO Dario Amodei argued that today’s technology is not yet reliable enough to hand critical lethal decisions entirely to autonomous AI systems.
The Pentagon took a different position, insisting that military commanders—not a private technology company—must ultimately control how legally acquired technology can be used in lawful military operations.
That disagreement exploded publicly earlier this year.
Hegseth announced that Anthropic would be treated as a supply-chain risk and said contractors doing business with the military would face restrictions involving Anthropic.
President Donald Trump also directed federal agencies to stop using Anthropic technology.
Anthropic sued.
By March, a federal court had already issued a preliminary injunction preventing agencies from implementing key portions of those measures while the litigation continued. The August 27 ruling subsequently delivered Anthropic an even larger victory on the merits of its claims.
Then Washington Suddenly Appeared to Warm Up to Anthropic
The political picture appeared to shift dramatically this week.
Lutnick told Axios that Anthropic had “done what we asked,” declaring that the company was again on the administration’s “right side.”
Anthropic co-founder Tom Brown also appeared alongside government and technology leaders at the G20 Innovation Ministerial in North Carolina, where he highlighted the economic importance of rapidly expanding AI data-center infrastructure.
Brown publicly praised Trump’s support for data-center development, another sign that relations between at least some parts of the administration and Anthropic had improved.
Then came Michael’s statement.
The Pentagon technology official made clear that, from the Defense Department’s perspective, Anthropic was still considered a supply-chain risk.
That means the apparent reconciliation in Washington may be far less complete than Lutnick’s comments suggested.
Is Anthropic Actually Blacklisted Right Now?
This is where the situation becomes legally complicated.
The Pentagon may continue to describe Anthropic as a supply-chain risk, but the federal court has ruled that the challenged designation and related government measures were unlawful.
The court had already prevented enforcement of key parts of the government’s actions through an injunction, and Lin’s August ruling found serious substantive and procedural violations in how the designation was imposed.
The Pentagon therefore retains broad authority to decide which AI companies it wants to contract with, but the court ruled that it cannot use an unlawful supply-chain designation or retaliatory government action to punish Anthropic for publicly disagreeing with military AI policy.
Anthropic, meanwhile, welcomed the ruling and said it remains interested in working with the government on national security.
“We remain focused on working productively with the government,” an Anthropic spokesperson said following the decision.
Why This Fight Is Bigger Than Anthropic
The dispute could become an important test case for the entire AI industry.
Frontier AI companies including Anthropic, OpenAI, Google and others are increasingly developing models powerful enough to play significant roles in cybersecurity, intelligence analysis, military planning and potentially autonomous systems.
That creates a conflict that Washington has not fully resolved.
Governments want access to the most powerful models available.
AI developers want safeguards against applications they believe their systems cannot safely perform.
And military officials are reluctant to allow private companies to exercise what could amount to a veto over battlefield technology.
Anthropic’s confrontation with the Pentagon has pushed that debate from theoretical AI-policy discussions into federal courts and high-level government contracting.
For now, Anthropic has won a significant legal victory.
It also appears to have repaired at least part of its political relationship with the Trump administration.
But Michael’s latest declaration makes one thing clear:
The Pentagon has not necessarily changed its mind about Anthropic—even if a federal judge has ruled that the government cannot treat that belief as legal justification for the sweeping blacklist it attempted to impose.
And that unresolved clash over who ultimately controls the rules of military AI could matter far beyond Claude.
WWC ONE MEDIA MJE
Leave a Reply