Anthropic counters Pentagon's national security risk claim in lawsuit response

Anthropic’s filing, obtained by The Economic Times, argues that the Pentagon’s allegation rests on a “mischaracterization of the company’s risk profile” and that the agency has not presented concrete evidence linking any of Anthropic’s deployed models to classified data leakage. The response cites internal audits that, according to the company, show “no unauthorized access or exfiltration” of government‑grade information from its Claude‑series systems, and it requests dismissal of the suit on procedural grounds.

The defense also highlights a parallel controversy involving Chinese AI firms that allegedly harvested Claude outputs to train their own models. Forbes reported that several Chinese startups created more than 24,000 fraudulent accounts to scrape Claude‑generated text, a claim echoed by VentureBeat, which named DeepSeek, Moonshot and MiniMax as the primary actors. Anthropic’s lawyers contend that the Pentagon’s national‑security concerns are being conflated with these unrelated intellectual‑property violations, noting that the Chinese activity “does not implicate the U.S. government’s own data handling practices” (Economic Times).

In addition to denying the security risk, Anthropic points to its compliance framework, which includes regular third‑party security assessments and a “zero‑trust architecture” for model serving. The company’s response cites specific technical safeguards—such as encrypted model weights at rest, isolated inference containers, and audit logs that record every API call—as evidence that its systems are “designed to prevent the kind of data exposure the Pentagon alleges.” These measures, the filing says, are consistent with industry best practices and have been validated by independent auditors, though the Economic Times does not disclose the auditors’ identities.

Anthropic also raises procedural objections, arguing that the Pentagon’s lawsuit was filed without first exhausting administrative remedies under the Department of Defense’s own procurement and security review processes. The filing claims that the department’s “premature legal action” undermines established channels for addressing alleged vulnerabilities and could set a precedent that hampers innovation across the broader AI ecosystem. The company urges the court to require the Pentagon to submit a detailed technical dossier of any alleged incidents before proceeding further.

Finally, the response underscores the broader competitive context, noting that the same Chinese firms accused of “mining” Claude data have been the subject of separate investigations by U.S. authorities. By separating the Pentagon’s national‑security claim from the intellectual‑property dispute, Anthropic seeks to isolate the lawsuit from the ongoing geopolitical tension over AI model theft, a narrative that has been amplified in recent coverage by The Information and VentureBeat. The company concludes that, absent specific, verifiable evidence of a security breach, the suit lacks merit and should be dismissed.