US Department of Defence stops using AI tools from Anthropic

The US Department of Defence has officially stopped using AI tools provided by Anthropic, a decision that comes months after the company was blacklisted as a supply chain risk on national security grounds.

According to a statement from the department, the Pentagon has ceased all use of Anthropic's products, with the move affecting various areas such as research, analysis, and intelligence gathering. The decision is reportedly linked to concerns about the potential risks associated with doing business with Anthropic.

The Pentagon first labelled Anthropic a supply chain risk in February, with Secretary of Defence Pete Hegseth announcing plans to stop using its tools by late August. However, it appears that there has been some delay in implementing this decision, with multiple sources confirming that Anthropic's AI model, Claude, was still being used by the department as recently as last week.

As a critical tool for processing large amounts of information, Claude has been viewed as essential to various military operations, including those against Iran. The use of Anthropic's tools in these areas is said to have been ongoing despite the initial announcement to cease using them.

The US government's relationship with Anthropic, a leading AI firm, has been strained since its designation as a "national security supply-chain risk" by the Pentagon earlier this year. This move was made after the company refused to remove safety guardrails from its AI product Claude and grant unfettered access to military agencies.

The designation is typically reserved for companies based in countries that pose a threat to US national security, but Anthropic's refusal to comply with the Pentagon's demands appears to be the reason behind this unprecedented move. The firm had been working with government agencies since 2024, making it the first advanced AI company to deploy its tools in classified work.

In response to the designation, Anthropic has taken legal action against the Trump administration, claiming that the move was both unlawful and retaliatory. This development comes as the Pentagon continues to sign contracts with other major tech companies, including Google, xAI, and OpenAI, suggesting a shift away from using Anthropic's services.

The use of these new AI tools by the military raises questions about their capabilities and potential applications, particularly in areas where Anthropic had previously been involved.

The adoption of OpenAI tools has been increasing in some military departments over recent months, according to individuals familiar with AI use within the defence department.

These new tools are being integrated into existing systems, including a larger data platform called Maven Smart System operated by Palantir for the Pentagon. The Maven system is primarily used for organising intelligence and other data across various departments.

The integration of OpenAI tools into the Maven system has made it difficult to remove them quickly from the system, according to Lauren Kahn, a senior research analyst at Georgetown's Center for Security and Emerging Technology. She noted that once these tools become deeply integrated, it can be challenging to remove them entirely.

Kahn, who is also a former US defence official, explained that this process of integration and removal is not a simple task, as evidenced by the Pentagon's delay in removing all use of Claude from its system.

The US government's decision to phase out Anthropic AI tools has raised concerns about finding alternative solutions for sensitive tasks like intelligence gathering and analysis.

A six-month deadline was set on 27 February for the complete removal of Anthropic from government systems, but employees continued to use its powerful Claude tools throughout the controversy surrounding the company.

These tools, including the Mythos models, are still being used in various capacities, particularly in conjunction with Palantir's Maven system.

Maven is a platform that allows analysts and defence officials to access and organise visual data, such as satellite imagery and drone footage, which is then fed into Claude and other language models for further analysis.

The Pentagon's decision to stop using Anthropic AI tools follows a recent blacklisting of the company by US authorities. This development comes as relations between the White House and Anthropic appear to be improving, with Trump meeting with the company's boss Dario Amodei twice at the White House last week.

After one meeting, Amodei expressed concerns about the risks posed by AI, stating that "how we address those risks is still under discussion" while also acknowledging the potential for safe collaboration with the president. The meetings were seen as an opportunity for Anthropic to win over support from Trump and his administration, with Trump praising Amodei as "great" and "fantastic".