Palantir, Nvidia and Booz Allen Reportedly Curb Frontier AI Over Data-Security Fears

Palantir, Nvidia and Booz Allen Hamilton are reportedly limiting or reconsidering use of advanced Anthropic and OpenAI models unless stronger guarantees are provided over how sensitive customer data is handled.

Reuters, citing The Information and people familiar with the discussions, reports that Palantir has demanded an irrevocable zero-data-retention guarantee from Anthropic before broadly offering its models through Palantir software.

Nvidia is reportedly restricting Anthropic models to less-sensitive work and using its own Nemotron systems for more sensitive tasks.

Booz Allen is said to have prohibited Anthropic’s commercial model for proprietary cybersecurity work.

Microsoft is pitching isolated cloud environments as an alternative.

Neither Reuters nor The Information established that Anthropic or OpenAI improperly used confidential material belonging to these companies.

The issue is trust.

The AI risk most companies are probably underestimating

Palantir, Nvidia and Booz Allen are not organisations unfamiliar with advanced technology.

They operate across defence, intelligence, infrastructure, cybersecurity and AI itself.

If companies with sophisticated security teams are demanding zero-retention guarantees before allowing frontier models near sensitive information, ordinary businesses should pay attention.

The danger is not necessarily that an AI provider is secretly misusing confidential data.

It is that employees may be feeding commercially sensitive information into systems without fully understanding where that data goes, how long it is retained or what contractual protections apply.

Businesses spent years training staff not to email confidential information to strangers.

Many are now pasting the same information into AI tools because the interface feels private.

It may not be.