Enterprise customers are putting data control ahead of frontier-model access

Reuters, relaying The Information’s reporting, says Palantir wants irrevocable zero-retention guarantees from Anthropic. Nvidia confines Anthropic usage to less-sensitive tasks, while Booz Allen bars its commercial model from proprietary cybersecurity work.

The dispute follows Anthropic’s June policy allowing 30-day usage-log retention to detect complex attacks. Both labs say they do not train on customer data by default. The companies did not immediately respond to Reuters’ requests for comment.

Retaining data for safety monitoring and using it for training are different questions; a promise about one does not settle the other.

My interpretation: enterprise model selection increasingly includes contractual and architectural constraints alongside quality and price. The strongest model may still be unsuitable for a workflow if the customer cannot accept where its inputs remain or who controls them. This is a commercial constraint on adoption, not evidence that customer secrets were actually misused.