Pentagon halts use of Anthropic AI tools following company’s blacklisting, BBC reports

KUALA LUMPUR, 6 OCTOBER 2026 —

The United States Department of Defence has officially ceased all use of Anthropic’s artificial intelligence tools, a decision that marks the culmination of a high-stakes conflict between the Pentagon and the leading AI developer.

This move follows a designation in February by Secretary of Defence Pete Hegseth, who labelled Anthropic a "supply chain risk" on national security grounds after the company refused to remove safety guardrails from its AI models.

The announcement resolves a significant ambiguity regarding the status of Anthropic’s Claude model within the US military infrastructure. For months, reports suggested that despite the official phase-out order, the technology remained deeply embedded in critical defence operations. The Pentagon’s confirmation that usage has stopped provides clarity on the extent of the government’s decoupling from the firm, a process that was complicated by the technical integration of the AI into the department’s primary data platforms.

The delay between the initial designation in February and the final cessation of use has raised questions about the practical difficulties of removing such deeply integrated systems from active military and intelligence workflows.

The friction between the Pentagon and Anthropic began earlier this year when the defence department pressed the company to remove safety guardrails from Claude, seeking unfettered access to its AI products for military applications. Anthropic refused, citing serious concerns regarding the potential for mass surveillance and the development of autonomous weapons.

In response, Hegseth designated the company a "national security supply-chain risk," a legal classification typically reserved for entities based in countries that pose a threat to the United States. Anthropic challenged this move, describing it as "unprecedented and unlawful," and subsequently sued the Trump administration in an effort to overturn the designation.

Despite the legal battle and the official phase-out order, multiple sources familiar with the matter indicated that Anthropic’s tools were still being used by the department as recently as last week. These individuals, who include former US defence officials and contractors who have worked closely with the Pentagon on AI initiatives, stated that the Claude model was actively employed in research, analysis, and intelligence gathering.

Crucially, the sources noted that the technology was also utilized in military operations against Iran, highlighting the operational dependency on the AI despite the political controversy surrounding its use.

The persistence of Anthropic’s tools within the Pentagon was attributed to their deep integration into the Maven Smart System, a larger data platform operated by Palantir. Maven serves as the Pentagon’s primary platform for organising intelligence and other data, making it a central hub for defence operations.

Lauren Kahn, a senior research analyst at Georgetown’s Center for Security and Emerging Technology and a former US defence official, explained that because Claude was deeply integrated into Maven, it would have been difficult to remove it quickly from the system.

She noted that the fact that it took the Pentagon until now to remove all use of Claude is indicative that these technologies are not "plug and play," adding that once integrated, they can be painful to remove.

The operational role of Claude within the Maven system was extensive. Sources familiar with the technology’s application said that analysts and other defence officials relied on Maven to access and organise visual data, including satellite imagery and drone footage. This data was then fed into Claude and other large language models for tasks such as identifying potential military targets.

According to a source who was given a demonstration of Maven by Palantir earlier this year, officials frequently used the technology when compiling one-page briefs that were passed up the chain of command. The system’s utility extended beyond the US, as Maven is also used by some US allies to share intelligence with the Trump administration.

The decision to phase out Anthropic was not without internal resistance. A former senior US defence official told the BBC that when Hegseth initially ordered Anthropic to be phased out of government, there was concern with moving beyond Anthropic to alternatives such as OpenAI and other tools.

The complexity of the transition was underscored by the timeline of events; Hegseth set a six-month deadline to phase out Anthropic on 27 February, one day before the US and Israel attacked Iran. Despite this deadline, Pentagon employees continued to use Anthropic’s Claude tools, including its powerful Mythos models, "throughout the whole period of the controversy" with the company, according to one of the people familiar with the matter.

While the Pentagon moved to sever ties with Anthropic, it simultaneously signed contracts with other major tech companies to ensure continuity in its AI capabilities. These new partnerships include agreements with Google, xAI, and OpenAI. The shift in vendor relationships reflects the broader strategic pivot of the US defence establishment towards a diversified portfolio of AI providers, reducing reliance on any single entity.

This diversification strategy was likely influenced by the legal and political uncertainties surrounding Anthropic’s designation as a supply chain risk, which created an environment of instability for government contractors and defence planners.

The legal and political landscape surrounding the dispute has been dynamic. As the lawsuit against the Trump administration played out in the courts, the Pentagon continued its operational use of Claude, creating a disconnect between legal status and practical application. The company’s refusal to compromise on safety guardrails, particularly regarding autonomous weapons and mass surveillance, stood in stark contrast to the military’s desire for unrestricted access.

This fundamental disagreement highlighted the growing tension between AI developers’ ethical frameworks and the operational demands of modern warfare, a tension that is likely to shape future policy debates on the use of AI in national security contexts.

Amidst the defence sector’s move away from Anthropic, relations between the company and the White House appear to be warming. President Donald Trump met with Anthropic boss Dario Amodei twice at the White House last week, a series of meetings that suggested a potential thaw in the political relationship. After one of the meetings with Trump and other technology leaders, Amodei addressed the ongoing risks posed by AI.

He stated that "how we address those risks is still under discussion," but added that "if we work with the president, we can win safely." Trump, standing by his side, praised Amodei, describing him as "great" and "fantastic."

The Pentagon’s statement on Monday came in direct response to a question from the BBC regarding its use of Anthropic, prompted by multiple interviews in which sources claimed Claude was still being widely used. The department’s confirmation that usage has ceased brings a definitive end to the operational phase of the dispute, even as the legal challenges continue.

The episode underscores the challenges of integrating advanced AI into critical national security infrastructure, where the removal of such tools is not merely a contractual matter but a complex technical and operational undertaking. The involvement of Palantir, founded by billionaire Peter Thiel, a known ally of President Trump, further illustrates the intricate web of relationships between tech leadership, political power, and defence procurement in the current administration.

The cessation of Anthropic’s use by the Pentagon marks a significant milestone in the evolving relationship between the US government and the AI industry. It demonstrates the willingness of the defence department to enforce supply chain designations, even against domestic tech giants, when safety and security concerns are raised. However, the delay in fully removing the technology highlights the practical limitations of such policies when applied to deeply embedded systems.

As the US military continues to rely on AI for intelligence and operations, the balance between security, safety, and operational efficiency will remain a central issue. The ongoing legal battle and the shifting political dynamics between Anthropic and the White House suggest that this dispute is far from over, with potential implications for the future regulation and deployment of AI in national security contexts.