Companies adopting advanced models are asking for two things that seem to be in tension: maintaining control over their data and preventing systems from being used for dangerous activities. Anthropic has announced Enterprise Frontier Safeguards, a solution developed with over a hundred customers to try to achieve both.
The system combines zero data retention with controls designed to detect misuse. The architectural difference is that data is stored within customer-controlled cloud infrastructure rather than by Anthropic. The rollout will begin in phases throughout the fall and will cover Claude Code, Claude Enterprise, the Claude platform, and environments distributed across AWS, Google Cloud, and Microsoft.
Privacy and security do not have to be mutually exclusive
Many regulated enterprises cannot simply send sensitive data to an external provider and leave it there. At the same time, a highly capable model without abuse prevention systems can increase operational risk. EFS therefore attempts to bring security controls close to the data, rather than requiring data to move toward the controls.
The solution was developed with companies across finance, healthcare, manufacturing, telecommunications, law, retail, and the public sector. It is a signal that enterprise AI is entering a phase where governance architecture matters almost as much as model quality.
The real market is trust infrastructure
Selling agents to enterprises requires more than just promising performance. It demands audits, data segregation, controls, logging, and accountability. Anthropic is trying to transform these requirements from an adoption hurdle into an integral part of the product.
