Anthropic Plans to Let Enterprise Customers Keep AI Data on Their Own Cloud

Anthropic is preparing to change its data-retention policy for enterprise customers using its most capable AI models. Under the planned system, businesses would still be required to retain data for 30 days for safety monitoring, but they would have the option of storing that information on their own cloud infrastructure rather than keeping it on Anthropic’s systems. The change is intended to give companies greater control over sensitive information while preserving Anthropic’s ability to monitor potential misuse.

The policy comes after Anthropic introduced 30-day retention for its most advanced models because the company believes some sophisticated misuse cannot be detected by examining individual requests. For example, potentially harmful activity could develop across multiple interactions, making longer-term analysis important for security teams. Anthropic’s latest safety report says the company considers this approach important for detecting coordinated cyber or biological threats.

The planned arrangement could be particularly attractive to companies in highly regulated industries. Keeping retained information inside a customer’s own cloud environment could give organizations greater control over sensitive business information, while still allowing Anthropic’s safety systems to operate under agreed security arrangements. Anthropic has reportedly been working with more than 100 customers, including Salesforce, while developing the system.

The announcement also comes amid growing competition over enterprise AI privacy. OpenAI recently introduced Private Safety Processing, which aims to detect misuse without retaining customers’ underlying data. That creates a direct contrast between the companies: Anthropic currently argues that some data retention is necessary for advanced safety monitoring, while OpenAI is attempting to achieve similar monitoring with a zero-retention approach.

For businesses, the issue goes beyond privacy. Companies increasingly want to deploy powerful AI models while maintaining control over proprietary information, customer data and regulatory compliance. Anthropic’s planned change could make its most capable models easier for security-conscious organizations to adopt. It also highlights a broader challenge for the AI industry: balancing increasingly powerful models with the need to monitor misuse without forcing customers to surrender control of sensitive data.