Pentagon stops using Anthropic AI tools months after blacklisting company, BBC told

* The US Department of Defense stopped using Anthropic products.
* The Secretary of Defense labeled Anthropic a supply chain risk in February.
* The Pentagon announced it would stop using Anthropic by late August.
* Anthropic had been used by the US government and military since 2024, deploying tools for classified work.
* The Pentagon pressed Anthropic to remove safety guardrails from Claude and grant unfettered access to AI tools earlier this year.
* Anthropic refused the request, citing concerns over mass surveillance and autonomous weapons.
* Anthropic sued the Trump administration regarding the designation.
* Claude was used for intelligence gathering and analysis, including through Palantir's Maven system.
* Maven is the Pentagon's primary platform for organizing intelligence.
* Pentagon employees continued to use Anthropic's Claude tools, including Mythos models, during the controversy.
* The Pentagon signed contracts with other tech companies, including Google, xAI, and OpenAI.

The narrative reveals a tension between executive directives, legal dispute, and the practical reality of deeply embedded technological systems within government structures. The reported cessation of use by the Pentagon contrasts sharply with unconfirmed reports suggesting continued operational use for sensitive tasks like intelligence gathering and military operations. This creates a gap between official policy and actual practice, highlighting the difficulty in enforcing centralized control over proprietary or deeply integrated technologies within large bureaucratic systems.
The fact that Claude was embedded within Maven Smart System, which is central to the Pentagon's intelligence organization, suggests that removing the tool requires systemic overhaul rather than simple deactivation. This points to a pattern where technological integration creates inertia; once a powerful system is woven into core operational platforms, disentangling it becomes complex, regardless of external mandates. The involvement of former defense officials and contractors underscores that these decisions are not purely technical but involve institutional knowledge, adding another layer of complexity to the "plug and play" critique regarding AI deployment.
The interaction between the legal fight against Anthropic and the subsequent engagement with other vendors like Google and OpenAI suggests a broader strategic shift where reliance on AI capabilities is being distributed across competing entities, irrespective of internal governmental restrictions. The final exchange between political figures suggests that technological risk management often operates in a space where public posturing and private negotiations regarding risk acceptance supersede strict adherence to immediate operational mandates.
Bridge Questions: If an organization like Maven is integrated with an external AI model, what specific structural changes are necessary to ensure real-time compliance when oversight demands mandate removal? How do legal liabilities for supply chain risks evolve when usage spans multiple jurisdictions and technologies? What institutional safeguards can be developed to manage technological risk effectively when operational necessity conflicts with security protocols?