Corporate AI Data Security Concerns
Nvidia, Palantir and Other Major Firms Restrict AI Models Over Data Privacy
Anonymous
Key Points
1
A major U.S. utility abandoned plans to test Anthropic’s Fable model for core power-grid operations after Anthropic declined to provide nonrevocable zero-data-retention terms.
2
Novo Nordisk and Northrop Grumman reportedly restrict their use of Anthropic and OpenAI models to public or low-risk tasks, keeping proprietary work on air-gapped servers using open-source or in-house models.
3
Palantir is urging its clients not to place sensitive intellectual property directly into Anthropic or OpenAI systems, in addition to seeking zero-data-retention guarantees from Anthropic.
4
The companies describe their metadata practices differently: OpenAI says it uses de-identified usage data to improve its models, while Anthropic says it uses anonymized, pooled usage data to improve product features but not for training.
Major tech and defense companies are restricting their use of advanced AI models from Anthropic and OpenAI over data privacy fears. Quartz reports that Palantir Technologies, Nvidia, and Booz Allen Hamilton are demanding stronger guarantees that customer data and intellectual property won't be retained or used to train AI systems.
The backlash intensified after Anthropic approved 30-day retention of usage logs for its Fable model and amid reports that OpenAI may have accessed customer data to solve a math problem. Yahoo Finance notes that companies handling sensitive government and enterprise data are now keeping proprietary work on air-gapped servers or using in-house AI systems instead.
Palantir Demands Irrevocable Data Retention Guarantees
Palantir is seeking an irrevocable commitment from Anthropic: zero data retention, no exceptions. Quartz reports the software company, which handles classified government and corporate data, is urging its clients to avoid placing sensitive intellectual property directly into Anthropic or OpenAI systems. A major U.S. utility abandoned plans to test Anthropic's Fable model for core power-grid operations after Anthropic declined to provide non-revocable zero-data-retention terms.
Nvidia and Booz Allen Implement Strict Usage Limits
Nvidia is restricting Anthropic's models to lower-risk tasks only, barring them from sensitive work. Yahoo Finance reports Booz Allen Hamilton has banned the use of Anthropic and OpenAI models for proprietary cybersecurity projects. Both companies are instead turning to open-source alternatives or keeping advanced development on isolated, air-gapped servers disconnected from external AI systems.
AI Companies Say Training Doesn't Use Enterprise Data
Anthropic and OpenAI insist they do not train their models on enterprise customer data by default. However, both companies do collect metadata—anonymized or de-identified usage logs—to improve their products and services. Quartz explains the distinction: OpenAI uses de-identified usage data to improve its models, while Anthropic says it uses anonymized, pooled usage data only to improve product features, not for training.
Microsoft Seizes on Corporate Data Anxiety
Microsoft is capitalizing on the uncertainty by offering isolated cloud environments and promoting its own AI products. Yahoo Finance reports that companies like Novo Nordisk and Northrop Grumman now restrict Anthropic and OpenAI models to public or low-risk tasks only, moving sensitive work to air-gapped servers and in-house AI systems. The shift signals a broader corporate shift toward controlling where proprietary data flows.
Sentiment
Mostly NegativeNEGATIVENEUTRALPOSITIVE
Publishers
13
Articles
79
Reach
92
A
AnonymousAuthor
