Why did Anthropic move to 30-day data retention for Claude Fable 5?

Anthropic says the most sophisticated misuse — stolen enterprise credentials, agents autonomously taking destructive actions, coordinated attacks spread across many sessions and accounts — cannot be detected by analysing each interaction in isolation and discarding it. Correlating across time and accounts requires storing data for a meaningful period, so Anthropic introduced 30-day retention starting with Fable 5. The company says it has never trained on enterprise data without explicit permission.

How much does Enterprise Frontier Safeguards cost?

Anthropic does not charge for EFS, and enabling it does not change model behaviour, API pricing or rate limits. Customers who keep monitoring data in their own cloud account pay their provider's normal storage, read, write and egress fees. Customer-owned storage, customer-managed encryption keys and fully automated review are each opt-in.

When is EFS available and on which platforms?

EFS rolls out in phases with broad availability targeted for later in fall 2026. It will be supported on Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Claude Platform on AWS, Google's Agent Platform and Microsoft Foundry, with equivalent controls whether you buy direct or through a cloud partner. Until their turn comes, eligible customers get zero data retention on Claude Fable 5 and Fable 5.1.

Quick Answer

What Is Anthropic Enterprise Frontier Safeguards (EFS)?

Q: What is Anthropic Enterprise Frontier Safeguards?

Enterprise Frontier Safeguards (EFS) is an Anthropic deployment option announced September 1, 2026 that combines zero-data-retention privacy with cross-session misuse detection. Activity data used for monitoring is stored in the customer's own cloud account — Amazon S3, Azure Blob Storage or Google Cloud Storage — under the customer's encryption keys, access policies and audit logging. Automated systems scan a rolling window of traffic for serious misuse and send flags directly to the customer's security team; no Anthropic employee reviews the content.

Published: September 9, 2026

The Short Answer

Enterprise Frontier Safeguards (EFS) is Anthropic’s answer to the data-retention dilemma it created with Claude Fable 5: the logs needed for misuse detection stay in your cloud account, under your keys, and the only humans who ever read a flagged alert are your security staff. Anthropic announced it on September 1, 2026, after designing it with more than 100 customers — a quarter of the Fortune 100 and every US global systemically important bank among them.

	Before EFS (Fable 5, July 2026)	With EFS (rolling out fall 2026)
Retention	30 days on Anthropic infrastructure	Rolling window in customer’s S3 / Azure Blob / GCS
Encryption keys	Anthropic-managed	Customer-managed (opt-in)
Who reviews flags	Anthropic automated + potential human review	Customer’s own team; no Anthropic human review
Price	—	Free; you pay your cloud provider for storage/egress
Model behaviour, pricing, limits	—	Unchanged

Why It Exists

Zero data retention (ZDR) — prompts and responses analysed and discarded — was the arrangement regulated buyers wanted. Anthropic dropped it for its most capable tier starting with Fable 5 and moved to a 30-day retention window, arguing that the most sophisticated misuse only shows up across many tasks, sessions and accounts. A stolen API key produces requests that look ordinary one at a time; only a rolling view catches the abnormal pattern.

Enterprises understood the security argument and still could not sign. Adding Anthropic as another “trusted data vendor” meant notifying their own customers, reworking contracts, and satisfying internal rules on who may see privileged legal material, non-public financial information, or drug-safety reports. The objection was never to review — it was to the reviewer.

EFS resolves that by splitting the roles: Anthropic operates the detection logic, the customer holds the data and staffs the alert queue.

How EFS Works

Customer-owned storage. Activity data used for monitoring is written to the customer’s own Amazon S3, Azure Blob Storage or Google Cloud Storage account, governed by the customer’s encryption keys, access policies and audit logging.
Automated monitoring. Anthropic’s systems analyse a rolling window of traffic for signals of serious misuse — attempts to develop offensive cyber or biological capabilities, and signs of stolen or leaked credentials. Anthropic has not published how long the window runs.
Flags go to the customer. When a pattern needs attention, the signal is sent directly to the customer’s security team. No human review by Anthropic employees is required.
Equivalent controls everywhere. The same design applies whether you buy from Anthropic directly or through AWS, Google Cloud or Microsoft Azure, with data landing in the cloud account you already trust.

Each of the three pillars — customer-owned storage, Customer-Managed Encryption Keys, and fully automated review — is opt-in, so an organisation can enable only what its regulators require.

Who Shaped It

Anthropic worked with the Analysis and Resilience Center for Systemic Risk (ARC), whose members include the CISOs of Goldman Sachs, Morgan Stanley, Citi, Bank of America and Wells Fargo, plus Comcast, KPMG, Mastercard, Salesforce and Visa. Wells Fargo’s CISO Munish Kumar Sharma described the outcome as “our logs stay in a Wells-managed environment under Wells-managed keys.” Stripe, Snowflake, FIS, Cognition and Factory have publicly said they will use it.

Availability, Cost and the Bridge Period

Rollout: phased, with broad availability targeted for later in fall 2026; access is by request form.
Platforms: Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Claude Platform on AWS, Google’s Agent Platform, Microsoft Foundry.
Cost: Anthropic charges nothing. Cloud storage, reads, writes and egress are billed by your provider at normal rates. Model pricing (Fable 5.1 at $10/$50 per MTok, cache reads $0.25) is unaffected.
Bridge: eligible customers receive ZDR on Fable 5 and Fable 5.1 until EFS is ready for them.

How It Compares

OpenAI previewed Private Safety Processing on August 19, 2026 with a similar goal — cross-interaction misuse detection without personnel access — but a different default: content stays on customer infrastructure for ZDR deployments, or on OpenAI storage encrypted with customer-held keys. EFS is more explicit about the alert queue being the customer’s to staff. See the side-by-side comparison for regulated buyers.