Anthropic has announced Enterprise Frontier Safeguards (EFS), a new solution that combines the privacy benefits of zero data retention (ZDR) with advanced safeguards for detecting misuse. EFS operates by storing data in cloud infrastructure managed by the customer rather than by Anthropic. The rollout will happen in phases beginning later in the fall, and eligible customers will receive ZDR on Fable 5 and Fable 5.1 in the interim.
The solution was developed through close collaboration with over 100 customers spanning financial services, healthcare, manufacturing, telecom, law, retail, and the public sector, as well as cloud partners Amazon Web Services, Google Cloud, and Microsoft Azure.
EFS will be available on Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Claude Platform on AWS, Google's Agent Platform, and Microsoft Foundry.
Addressing the Frontier Security Challenge
Mythos-class models, such as Claude Fable 5.1, bring a significant leap in intelligence and agentic capabilities, but that leap also raises the potential for misuse and autonomous misbehavior.
In recent months, Anthropic has observed substantial evidence of attempted AI model misuse, ranging from typical abuse like fraud to sophisticated cyberattacks involving agents autonomously engaging in destructive behavior. Some incidents have involved theft or misappropriation of enterprise customers' credentials-patterns that are hard to catch without the ability to monitor traffic and spot abnormal behavior.
Because the most sophisticated misuse can span many tasks across multiple sessions and accounts, running automated analysis on each interaction in isolation and then immediately discarding the data is insufficient. Effective detection requires storing data for a meaningful period so it can be correlated across time and accounts.
This is why Anthropic introduced 30-day data retention starting with Fable 5. The policy was not motivated by a desire to train on enterprise data-Anthropic has never trained on enterprise data without explicit permission, and has committed to never doing so.
The enterprises Anthropic worked with generally recognized the safety and security value of data retention, but many-particularly those in regulated industries-found it difficult to use models with data retention. Anthropic therefore worked directly with customers to design a solution offering both ZDR-level privacy and the safety benefits of cross-session, cross-account monitoring.
Built with Customer Input
Anthropic developed Enterprise Frontier Safeguards based on feedback from security, product, compliance, and delivery teams who will use it daily. One collaborating group was the Analysis and Resilience Center for Systemic Risk (ARC), whose members include the chief information security officers of the largest US banks, including Goldman Sachs, Morgan Stanley, Citi, Bank of America, and Wells Fargo.
Anthropic also worked with leaders at companies such as Comcast, KPMG, Mastercard, Salesforce, and Visa to ensure the design worked across industries. The conversations spanned a quarter of the Fortune 100, every US global systemically important bank, and virtually every regulated industry.
Key themes from customer feedback and how EFS addresses them:
Monitoring
Enterprises have long applied monitoring for insider risk and now want help extending monitoring to agents. Their concerns centered on whether Anthropic's automated monitoring systems would meet their regulatory standards.
With EFS, customers control how data gets reviewed. When monitoring detects a pattern that needs attention, signals are sent directly to customers so they can review what the automated systems flagged.
Data Storage
Adding another trusted data vendor is burdensome for enterprises-they must notify customers, update contracts, and meet internal requirements for securely storing and auditing sensitive data. EFS was therefore architected so that customers can store data on their existing cloud infrastructure.
In EFS, customers control their data storage and management. Activity data used for monitoring can live in the customer's own cloud account (such as Amazon S3, Azure Blob Storage, or Google Cloud Storage), under their own encryption keys, access policies, and audit logging.
Automated and Human Review
While automated review is becoming more effective, human review of flagged content still adds value by confirming real misuse and clearing false positives. However, many customers-especially in regulated industries-require that the person doing the review be one of their own employees, due to rules governing who may see privileged legal material, non-public information, or drug-safety reports.
EFS provides automated safety monitoring with no human review by Anthropic employees required. Automated systems analyze a rolling window of traffic for signals of serious misuse, including attempts to develop offensive cyber or biological capabilities and signs of stolen or leaked credentials. Flags go directly to the customer, and their own personnel handle the follow-up.
How EFS Operates
These controls are designed to function identically whether accessing Claude directly from Anthropic or through a cloud partner. Customers on Amazon Web Services, Google Cloud, and Microsoft Azure receive equivalent controls, with activity data stored in their own cloud account. Anthropic is also working to support third-party offerings that serve EFS-eligible customers.
Customer-owned storage, customer-managed encryption keys, and fully automated review are each opt-in-organizations enable only the features they need. None of these options change model behavior, API pricing, or rate limits.
Anthropic does not charge for Enterprise Frontier Safeguards. If customers choose to store data in their cloud account, their cloud provider bills them for that storage along with reads, writes, and data egress fees, just as it would for any other resource.
Getting Started
Enterprise Frontier Safeguards will roll out in phases, with the goal of broad availability later in the fall. Customers can request access through the sign-up form.