Nvidia, Palantir Restrict Advanced AI Model Usage Over Rising Data Privacy Concerns

Major Tech Firms Push Back on AI Model Providers

Data privacy concerns are forcing some of the world's largest technology companies to restrict advanced AI model usage from providers like Anthropic and OpenAI1

1

2

2

. Palantir and Nvidia have placed significant limitations on which models their employees can access, while Booz Allen has barred staff from using Anthropic's commercial model for proprietary cybersecurity projects3

3

. The core issue centers on intellectual property risks and fears that sensitive corporate data could be retained, analyzed, or used to train future models—potentially handing competitors valuable business intelligence.

Palantir has pressed Anthropic to provide irrevocable zero data retention guarantees before making its models available through Palantir's software2

2

. Nvidia limits Anthropic's use to less sensitive tasks and relies on its own Nemotron models for internal work4

4

. At least one large U.S. utility company has already canceled plans to test Anthropic's Fable model after the company refused to agree to a nonrevocable zero data retention policy1

1

.

The Fable Model Policy Change Triggers Customer Exodus

The controversy traces back to a June policy change by Anthropic for its flagship Fable model1

1

. Following the update, Anthropic gained rights to retain customer usage logs for 30 days to guard against "complex and novel attacks"4

4

. While Anthropic argues it only retains data to ensure the Fable model isn't being misused, companies worry that sensitive business data will be caught in the sweep.

Both OpenAI and Anthropic maintain they don't train their models on information provided by companies with specific enterprise contracts by default1

1

. However, both companies do collect metadata from corporate customers. OpenAI notes this metadata is used "to better understand how our services are used," while Anthropic claims any collected data is aggregated and anonymized1

1

. Neither company uses this metadata to train models, they say.

What Companies Fear About Third-Party AI Model Risks

Telecoms company C Spire has agreements with both OpenAI and Anthropic preventing either from using its data to train models1

1

. Yet the contracts still allow both companies to collect technical usage data. C Spire believes this includes information about what applications AI models connect to, usage patterns, and potentially what models process between generating responses—so-called "chain-of-thought" data. OpenAI states it doesn't use this intermediate reasoning data to train models, but C Spire argues neither AI company provides clear explanations about exactly what gets collected1

1

.

David Friedberg, CEO of agricultural biotechnology company Ohalo, described moving off Anthropic's models for sensitive corporate data and shifting toward locally run models5

5

. Friedberg recounted instances where novel scientific ideas discussed with one model appeared to resurface when he queried another version from a different account5

5

. His concern extends beyond leaked documents to the possibility of handing AI providers valuable proprietary know-how developed inside a company.

Air-Gapped Servers and Isolated Cloud Environments Emerge as Solutions

Aerospace company Northrop Grumman has chosen to run open-source AI models on its own air-gapped servers rather than trusting external providers like OpenAI and Anthropic1

1

. Pharmaceutical giant Novo Nordisk continues using Anthropic's models for some tasks but maintains a strict ban on allowing any proprietary data to interact with the model1

1

.

Microsoft is capitalizing on these data privacy concerns by pitching isolated cloud environments and its own AI offerings to draw clients away from Anthropic and OpenAI2

2

4

4

. Microsoft's isolated cloud environments run AI models on private servers that don't transmit data to external AI companies, though this approach proves costly1

1

.

Nvidia and Palantir recently unveiled a joint system using Palantir software and Nvidia's open Nemotron models to analyze supply-chain data5

5

. Nvidia serves as the first customer for this technology, which allows companies to customize models using proprietary operational data while maintaining full control over where the system runs. For Nvidia, this means capturing how its planners make decisions without exposing supply-chain knowledge to outside model providers5

5

.

Anthropic's Response and Industry Implications

Anthropic has acknowledged that many enterprises, particularly in regulated industries, find the 30-day retention requirement difficult to accept5

5

. The company's proposed solution, Enterprise Frontier Safeguards, will keep activity data used for misuse monitoring within customer-controlled cloud infrastructure. The feature begins rolling out in phases later this fall5

5

.

Despite these challenges, market expectations for Anthropic remain strong. The lack of trust over data handling has already cost AI companies real business, but Anthropic CEO Dario Amodei recently called for AI companies to slow frontier model development, warning that capabilities advance faster than researchers can understand or control—a position endorsed by OpenAI CEO Sam Altman2

2

4

4

.

Watch for how enterprise customers negotiate data handling terms in coming months. Companies investing heavily in AI infrastructure may accelerate shifts toward self-hosted models or demand contractual guarantees that eliminate any ambiguity about data retention. The tension between AI providers needing usage data to improve safety and customers protecting competitive advantage will likely reshape enterprise AI contracts industry-wide.