US agencies drop Anthropic Claude amid cloud and data risk worries

In short: A growing “cloud of doubt” is forming as governments and companies worry that AI tools can expose sensitive data and create risky dependence on cloud services.

What's going on

Concerns are rising about what happens when AI features are plugged into everyday work tools and cloud services (computing that runs on someone else’s servers over the internet). When more work moves into these systems, it can become harder to control who sees the data, how fast systems respond, and what happens if the internet or a provider goes down.

A major example is a dispute between Anthropic and the US Pentagon. Reports say Anthropic refused requests that crossed its “red lines,” including using its Claude AI for mass surveillance of US citizens or for autonomous weapons. After talks collapsed, the Pentagon labeled Anthropic a “supply chain risk” and federal agencies were directed to stop using it, switching to models from OpenAI, Google, and xAI.

Legal experts have questioned whether the administration can enforce such a broad ban, especially when many contractors use Claude inside larger products from companies like Google, Microsoft, and Nvidia. Claude was also reported to have been cleared for some classified work, which made the split even more disruptive.

Separate research has added to the unease. A joint study by ETH Zurich, Anthropic, and MATS reported that large language models can sometimes identify pseudonymous people by comparing what they said in interviews with public posts on sites like Reddit and LinkedIn. Think of it like matching a “nameless” quote to a person by recognizing their writing habits.

What to watch

Watch whether more agencies and regulated industries demand “on-premise” options (running AI on their own computers, like keeping files in a home safe instead of a storage unit). Also watch for tighter rules on which AI models can be used, and where their data is allowed to travel.

Source: NYTimes