On September 14, The Information reported that NVIDIA, Palantir, and Booz Allen Hamilton have started to impose restrictions on the internal use of Anthropic's AI models. The primary concern driving this move is the potential risk of employees inputting sensitive information, such as internal code and intellectual property, into external AI models, which could inadvertently grant the model providers access to or even enable them to learn from such data. Specifically, NVIDIA restricts the use of Anthropic's Fable to tasks with lower sensitivity levels, such as open-source software development. For more sensitive internal projects, like supply chain monitoring, the company relies on its proprietary Nemotron model. Palantir, on the other hand, mandates that Anthropic provides assurances of "irrevocable zero data retention" before it makes the relevant models widely available to customers through its own software platform. Meanwhile, Booz Allen Hamilton prohibits its employees from using Anthropic's commercial models for handling proprietary cybersecurity tasks. In stark contrast, Google permits all its engineers to utilize Anthropic's Claude model without such restrictions.