Archive · 15 Sep 2026
Palantir, Nvidia May Restrict AI Over Data Safety Fears
Some large companies are worried about how advanced AI systems handle private information and computer code.
Reports say Palantir, NVIDIA and Booz Allen Hamilton may limit their use of certain AI models.
Palantir reportedly wants Anthropic to promise that it will never keep customer data.
NVIDIA is using Anthropic models only for less-sensitive work and relies on its own models internally.
Booz Allen has stopped employees from using Anthropic’s commercial model for certain cybersecurity projects.
Microsoft is offering private cloud systems and its own AI tools to companies with these concerns.
Anthropic and OpenAI say they do not use customer data to train models unless customers agree.
Anthropic’s chief executive has also urged AI companies to slow the development of very advanced models, and other technology leaders supported that idea.
Palantir Technologies, NVIDIA Corporation and Booz Allen Hamilton may restrict or stop using advanced AI models over intellectual-property and data-safety concerns.
Palantir has reportedly sought an irrevocable zero-data-retention guarantee from Anthropic before offering its models through Palantir software.
NVIDIA reportedly limits Anthropic models to less-sensitive tasks and uses its own Nemotron models internally, while Booz Allen restricts their use for proprietary cybersecurity work.
Microsoft is promoting isolated cloud environments and its own AI services to attract customers concerned about data protection.
Anthropic and OpenAI say they do not train on customer data by default, although they collect anonymized metadata to improve their products.
- Who
- Palantir Technologies, NVIDIA Corporation, Booz Allen Hamilton, Anthropic, OpenAI and Microsoft Corporation are central to the report; executives Dario Amodei, Sam Altman and Elon Musk also made relevant comments.
- What
- Several companies may limit or stop using advanced AI models because of concerns about data retention, intellectual-property misuse and the pace of AI development.
- When
- The report was published Monday; Dario Amodei made his call to slow frontier-model development on Saturday, and Anthropic changed its Fable policy in June.
- Where
- The reports do not specify a particular location.
- Why
- Companies are concerned about potential misuse of intellectual property, retention of usage logs and possible use of customer data in model development.
Corporate caution
AI labs and providers
Customer-data protection
Corporate caution
Companies are seeking stronger guarantees, including Palantir’s reported demand for irrevocable zero-data-retention protections, and are limiting AI use for sensitive work.
AI labs and providers
Anthropic and OpenAI say they do not train on customer data by default unless companies opt in, while acknowledging that anonymized metadata is collected to improve products.
Use of advanced models
Corporate caution
NVIDIA reportedly restricts Anthropic models to less-sensitive tasks, and Booz Allen bars their use for proprietary cybersecurity work because of data and intellectual-property concerns.
AI labs and providers
Microsoft is presenting isolated cloud environments and its own AI offerings as alternatives for customers concerned about how advanced models handle sensitive information.
Pace of AI development
Corporate caution
Anthropic CEO Dario Amodei argued that frontier-model development should slow because capabilities are advancing faster than researchers can understand or control.
AI labs and providers
OpenAI CEO Sam Altman and Elon Musk quickly endorsed Amodei’s call, while the reports do not describe a disagreement from those leaders.
- Companies considering restrictions
- Palantir Technologies, NVIDIA Corporation and Booz Allen Hamilton
- Palantir’s reported demand
- An irrevocable guarantee from Anthropic that it will retain no data
- NVIDIA’s reported approach
- Limiting Anthropic models to less-sensitive tasks and using Nemotron models internally
- Booz Allen’s reported restriction
- Employees are barred from using Anthropic’s commercial model for proprietary cybersecurity work
- Microsoft’s strategy
- Pitching isolated cloud environments and its own AI offerings
- AI labs’ stated policy
- Anthropic and OpenAI say they do not train on customer data by default unless companies opt in
- Anthropic policy change
- A June change to the Fable model policy allowed 30-day retention of usage logs to help guard against complex and novel attacks