Palantir and Nvidia Reportedly Limit Data Access for Anthropic Model Over Retention Concerns

Palantir Technologies and Nvidia have moved to restrict how Anthropic’s language model, referred to as “Fable,” can be used within their environments, according to a report by PYMNTS.com. The reported restriction centers on data retention practices, an increasingly contentious issue as defence and enterprise customers weigh the compliance risks of third-party AI models that may retain or learn from sensitive inputs.

The development highlights a growing tension in the defence technology ecosystem. Contractors and their suppliers handle large volumes of proprietary and government-related information, and many operate under strict contractual and regulatory obligations — including frameworks such as FedRAMP (the Federal Risk and Authorization Management Program) and ITAR (the International Traffic in Arms Regulations) — that govern where data lives, how long it is stored, and who can access it. A third-party model whose data retention policies do not align with those requirements can create friction, even when the underlying technology is otherwise attractive.

For Palantir, the report lands amid continued investor attention on its position in the AI software landscape. Shares of Palantir Technologies Inc. (PLTR) were trading at $170.50, down 0.07% from a previous close of $170.62, giving the company a market capitalization of roughly $439.7 billion. The Denver-based firm, which classifies within the Technology sector’s Software – Infrastructure industry, builds data platforms used by government and commercial customers, including its Gotham platform deployed by intelligence and defence organizations.

Restrictions of this kind also reflect a broader pattern in enterprise AI procurement: companies are increasingly segmenting their AI stacks, permitting certain models for general-purpose work while limiting or excluding others from pipelines that touch regulated or customer-sensitive data. Data retention terms — how long a provider stores prompts, outputs, and derived artifacts — have become a standard line item in vendor evaluations, alongside security certifications and deployment models such as on-premise or virtual private cloud hosting.

Neither the scope of the reported restrictions nor their commercial impact was detailed in the report. It remains unclear whether the limits apply to specific product lines, deployments, or internal use cases.

What to watch

• Any formal statement or policy update from Palantir, Nvidia, or Anthropic clarifying the nature and scope of the data retention dispute.
• Palantir’s upcoming earnings release and management commentary on AI partnerships and third-party model integration.
• Broader industry movement on AI vendor data retention standards, particularly among government-facing software providers.

Source: original release