America

Pentagon to fully purge Anthropic AI from systems by end of September

Published

on

The US Department of War plans to eliminate all technology belonging to artificial intelligence company Anthropic across its internal systems by the end of September.

Emil Michael, the Pentagon’s chief technology officer and Under Secretary of War for Research and Engineering, revealed in an interview with Breitbart that the department is systematically purging the software.

“We are phasing their software out of all department systems. Over 90% has been deleted so far. We will conclude this process by the end of September,” Michael stated.

Noting that Anthropic technologies are integrated into complex architectures, Michael explained that the removal process takes time and must be executed responsibly.

The dispute between the Pentagon and Anthropic originated from restrictions imposed by the company regarding military use.

Anthropic Chief Executive Officer Dario Amodei demanded assurances that the technology would not be used in autonomous weapons or surveillance activities without human oversight. Following this stance, the Pentagon designated Anthropic a “supply chain risk”. Nearly 100% of military systems have transitioned to alternative AI models, including the Maven system, which aided target acquisition for airstrikes during the war with Iran.

In May, The Wall Street Journal reported that Claude Mythos, an AI model developed by Anthropic to identify software vulnerabilities, caused chaos within the administration of US President Donald Trump.

Subsequently, Mark Warner, Vice Chairman of the US Senate Select Committee on Intelligence, reported that Anthropic’s AI model had breached classified US National Security Agency (NSA) systems.

Preliminary injunction hearing held in federal court

The first major hearing in the lawsuit challenging the Pentagon’s supply chain risk designation of Anthropic was heard Tuesday afternoon in federal court in San Francisco.

US District Judge Rita Lin met the Department of War’s decision with scepticism, implying the move could be retaliation intended to punish a company that opposed the Trump administration during contract negotiations over AI deployment.

During the hearing, Judge Lin remarked: “What troubles me about these reactions is that they do not seem entirely congruent with the stated national security concern. If the concern were about the integrity of the operational chain of command, they could simply stop using Claude. It appears the defendants went further and sought to punish Anthropic.”

Concluding the hearing without an immediate ruling, Judge Lin stated she would issue a decision on the preliminary injunction request within the coming days.

Lin observed that the effort to bar the company from all government business went beyond what was necessary to address the national security concern cited by Secretary of War Pete Hegseth, stating: “This appears to be an attempt to crush Anthropic, and my specific concern is whether Anthropic is being punished for publicly criticising the government’s contracting stance.”

While acknowledging Hegseth’s authority to restrict the company’s applications in the national security sphere, Lin emphasised that the core issue is whether the Pentagon exceeded its authority by designating the company a supply chain risk.

Government and company counsel clash over sabotage

Justice Department officials representing the Trump administration argued that Anthropic poses a national security risk due to concerns that it could “sabotage” the military via a future software update.

Acting Deputy Assistant Attorney General Eric Hamilton stated: “Anthropic is not merely being stubborn or refusing contract terms; instead, it is communicating concerns to the Department of War about how the technology is used even in military missions covered under terms of service.”

Criticising this rationale, Judge Lin noted it was an insufficient basis for barring a company from government work, asking: “If an IT supplier acts stubbornly, insists on specific contract terms, and raises uncomfortable questions, is that sufficient to declare it a supply chain risk on the grounds that it may be untrustworthy?”

When questioned on how the company could sabotage the military, Hamilton suggested Anthropic might attempt to introduce a “kill switch” into the software if it disapproved of how Claude was being utilised.

Hamilton could not definitively confirm whether the company possessed the capability to deploy such an update. “The concern is that, rather than merely stating concerns and lodging objections, Anthropic might take issue with the department’s actions and decide to manipulate the software to control its use by the Department of War,” Hamilton said.

During the proceedings, Hamilton effectively contradicted a social media post made last month by Hegseth, which asserted that as a result of the dispute, “no contractor, vendor, or partner doing business with the US military may conduct any commercial activity with Anthropic.”

Hamilton said he was unaware of any statutory authority permitting the department to categorically prohibit contractors from using a company’s products or services on non-departmental business.

When Judge Lin asked why Hegseth would make such an assertion if he did not intend it, Hamilton replied: “I do not know. What I do know is that this administration is committed to transparency and indicated in that post it would take steps to declare the company a supply chain risk, which it did.”

Anthropic’s attorney, Michael Mongan, argued that the Department of War’s positions shifted constantly as it sought to penalise the company following failed contract negotiations.

“The reality is that this is a supply chain designation in search of a justification or logic, and that constitutes a problem under the Administrative Procedure Act,” Mongan stated.

Requesting an immediate preliminary injunction to invalidate the Pentagon’s actions, Mongan said: “These actions are unlawful for the reasons we have discussed. They have inflicted immediate, irreparable, and ongoing harm on Anthropic, its constitutional rights, its reputation with partners, customers, and investors, and its broad commercial interests beyond the national security sector.”

MOST READ

Exit mobile version