Big Tech Wants AI Labs To Promise Hands-Off Data

Palantir, Nvidia, and Booz Allen are pushing Anthropic and OpenAI for tougher zero-retention guarantees as Microsoft pitches isolated cloud setups to win cautious clients.

What’s going on here?

Some of the biggest corporate AI buyers are telling OpenAI and Anthropic that “zero data retention” needs to be non-negotiable if they want access to sensitive work, The Information reported.

What does this mean?

The fight is about what happens to the prompts, files, and logs companies send into frontier models, especially when they include intellectual property or cybersecurity details. According to The Information, Anthropic faced pushback after a policy change that allowed it to keep certain usage logs for 30 days to help investigate sophisticated attacks, and OpenAI has faced scrutiny over whether user data was used in training for a specific test case. Both companies say they don’t train on customer data by default unless a client opts in, but they still collect some anonymized metadata, and for cautious clients that’s still a risk.

That tension is changing how AI gets deployed inside large organizations. Instead of a simple, shared application programming interface (API) where the vendor controls the service, buyers are leaning toward isolated setups in the cloud or on private infrastructure, where data access and log retention can be tightly controlled. The report says Microsoft is pitching these “walled-off” cloud environments alongside its AI products, while Palantir is asking for an irrevocable zero-retention promise before distributing Anthropic models, Nvidia is keeping Anthropic to less sensitive uses while using its own Nemotron models internally, and Booz Allen is restricting Anthropic’s commercial model for proprietary cybersecurity work.

Why should I care?

For markets: Microsoft can sell isolation as a paid feature while AI labs face tougher enterprise gatekeepers.

When security teams treat retention policies as a deal-breaker, it creates a procurement bottleneck that hits adoption before model quality does. That tends to favor companies that can bundle compliance controls, auditability, and segregated compute into a single enterprise package – which is where Microsoft’s cloud business has leverage. For frontier-model vendors like OpenAI and Anthropic, the pressure is to offer stricter, clearer guarantees or risk being limited to lower-sensitivity workloads, while the highest-stakes projects shift to private deployments or in-house models.

—

Originally Posted September 14, 2026 – Big Tech Wants AI Labs To Promise Hands-Off Data