Nvidia and Palantir Technologies are tightening controls around the use of Anthropic’s Claude models, reflecting growing concern among large enterprises over how sensitive corporate information is handled when it passes through frontier AI systems.
The restrictions are primarily about data security, intellectual property and retention, rather than fears about what employees might use AI to generate, according to reporting by The Information that was subsequently confirmed by Reuters. Nvidia is restricting Anthropic models to less-sensitive work, while Palantir has sought stronger contractual guarantees from Anthropic around zero data retention. The development comes as companies increasingly integrate powerful external AI models into coding, cybersecurity, research and internal operations, raising a central enterprise question: how much confidential data should businesses expose to third-party model providers?
Companies Seek Stronger Control Over Proprietary Data
Palantir has reportedly asked Anthropic for an irrevocable commitment that data submitted through its systems will not be retained, while Nvidia has limited Claude’s use in areas involving more sensitive information. Booz Allen Hamilton has also restricted employees from using Anthropic’s commercial AI products for proprietary cybersecurity work. The concerns have intensified following changes to Anthropic’s handling of certain model interactions. The company introduced retention of some user logs for up to 30 days as part of efforts to investigate sophisticated attacks and model misuse, according to The Information and Reuters. That has prompted some enterprise customers to seek additional contractual protections before putting highly confidential material through Claude.
The issue extends beyond Anthropic. Companies are also scrutinising OpenAI and other external AI providers over whether proprietary information could be stored, inspected or potentially contribute to model improvement. For businesses working with government information, cybersecurity systems, chip designs or other valuable intellectual property, even limited data retention can create significant security and compliance concerns.
Nvidia and Palantir Push Toward ‘Sovereign AI’
The restrictions also fit with a broader shift toward AI systems that allow companies to retain greater control over their own data and infrastructure. Just days earlier, Nvidia and Palantir announced a partnership centred on what they describe as “sovereign AI.” The companies are integrating Nvidia’s Nemotron open models with Palantir’s Foundry and Artificial Intelligence Platform, initially for Nvidia’s own supply-chain operations. Nvidia said the architecture is designed to let organisations retain control and ownership of proprietary data while deploying AI either in the cloud or on their own infrastructure. The approach highlights an emerging competitive divide in enterprise AI. The most capable closed models from companies such as Anthropic and OpenAI can offer superior performance for complex tasks, but running models externally can require organisations to place greater trust in the provider’s security and data-handling policies.
Open or privately deployed models can give businesses more direct control, even when their capabilities do not always match the most advanced frontier systems. For Anthropic, the restrictions do not amount to a broad rejection of Claude by Nvidia or Palantir. Instead, they illustrate how rapidly enterprise AI adoption is moving from questions about model capability toward questions of who controls the data, where it is stored and what happens to it after a model processes it.