Anthropic Launches Enterprise Frontier Safeguards to Shift Data Control to Customers
The new EFS framework allows companies to maintain physical control of their data while retaining AI misuse detection.
Anthropic has introduced Enterprise Frontier Safeguards (EFS), a new data handling framework that allows corporate clients to store AI-related data within their own cloud infrastructure. The move aims to resolve the inherent tension between the need for safety monitoring and the strict privacy requirements of large-scale enterprises.
According to the company, EFS combines Zero Data Retention (ZDR) with robust misuse detection capabilities. Under this new system, data is stored in cloud infrastructure controlled by the customer rather than by Anthropic. The solution is supported across Claude Code, Claude Enterprise, and the Claude Platform, as well as through major cloud partners including AWS, Google Cloud, and Microsoft Azure. Anthropic announced in September 2026 that the rollout of these safeguards will begin later this fall.
The Path to Decentralized Safety
The development of EFS was a collaborative effort involving more than 100 customers. These partners spanned a wide array of highly regulated sectors, including healthcare, financial services, law, telecom, manufacturing, retail, and the public sector. This broad collaboration reflects the specific challenges faced by industries where data residency and isolation are non-negotiable legal or operational requirements.
Historically, AI providers have faced a dilemma: safety monitoring typically requires the provider to have access to data to detect harmful use, but corporate privacy mandates that sensitive data remain isolated. By shifting the storage layer to the customer's environment, Anthropic is attempting to decouple the ability to monitor for misuse from the physical possession of the data.
Industry Implications
This shift toward decentralized safety monitoring sets a significant precedent for the frontier model industry. It suggests a future where AI labs provide the tools and logic for safeguards, but the customer retains absolute physical control of the underlying data. For Fortune 500 companies, this removes a primary barrier to integrating frontier models into core workflows, as it aligns AI adoption with existing corporate data governance policies.
Looking Ahead
As the rollout progresses this fall, the industry will be watching to see if other major AI labs adopt similar decentralized architectures to attract regulated industries. While the technical framework for EFS is now established, the effectiveness of misuse detection when data is stored externally remains a key point of interest for safety researchers and enterprise security officers alike.