Glonce

Anthropic ties zero data retention on Fable 5.1 to new abuse monitoring

securityThe Glonce editorsThe Register🔥64Glonce rating64Impact60Novelty45Relevance95Surprise35Players85Credibility75
Anthropic ties zero data retention on Fable 5.1 to new abuse monitoring

Anthropic has introduced Enterprise Frontier Safeguards (EFS), a program timed to the release of its new Fable 5.1 and Mythos 5.1 models that restores zero data retention (ZDR) for approved enterprise customers of Fable 5.1. Enterprises do not get ZDR automatically: they must apply and be approved. Without a special arrangement, the default for commercial API customers is 30 days of retention for prompts and outputs.

Earlier this year Anthropic began temporarily keeping inputs and outputs of Fable and Mythos, even for customers who already had ZDR agreements, so it could screen the material for safety violations. The company says it had seen "substantial evidence of attempted misuse of AI models" over recent months and needed to store data briefly to correlate events across accounts over time. Anthropic says this retained data is not used to train its models.

Enterprise customers pushed back regardless. Anthropic said the businesses it worked with generally understood the security value of retention, but many, especially in regulated industries, found it hard to use models that retained their data at all. Rival OpenAI appears to have run into the same pressure: it unveiled a comparable feature, Private Safety Processing, the month before this announcement, letting it flag risky use without breaking its own ZDR commitments.

Anthropic says EFS works by storing the relevant data in cloud infrastructure controlled by the customer, not by Anthropic. Automated abuse monitoring keeps running, but flagged activity goes straight to the customer instead of Anthropic's own staff: "When monitoring detects a pattern that needs attention, those signals are sent directly to customers so they can review what the automated systems detected." That shifts part of the safety-review workload from Anthropic onto the enterprise itself.

Anthropic plans to release EFS this fall, with no more specific date given. It will be offered by invitation for Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Claude Platform on AWS, Google's Agent Platform and Microsoft Foundry. Mythos stays under Anthropic's earlier retention policy and is not covered by EFS. Consumer plans, Claude Free, Pro and Max on the web, desktop and mobile, plus Claude.ai and Claude Code, continue under the company's previous data retention terms.

Key facts

  • Anthropic launches Enterprise Frontier Safeguards (EFS) alongside Fable 5.1 and Mythos 5.1, restoring opt-in zero data retention for approved Fable 5.1 enterprise customers.
  • ZDR is not automatic: enterprises must apply and be approved; the default retention for commercial API customers without a special arrangement is 30 days.
  • Anthropic says it began temporarily retaining Fable and Mythos data earlier this year after seeing "substantial evidence of attempted misuse of AI models," to correlate abuse across accounts, and says this data is not used for training.
  • EFS stores data in the customer's own cloud infrastructure rather than Anthropic's, and routes automated abuse-detection alerts directly to the customer to review, shifting some monitoring work onto them.
  • EFS is planned for this fall, by invitation, for Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Claude Platform on AWS, Google's Agent Platform and Microsoft Foundry; Mythos and consumer plans (Claude Free, Pro, Max, Claude.ai) keep the earlier policy.

Why it matters

Enterprise buyers, especially in regulated industries, have treated data retention as a reason to avoid a model provider, even when the provider explains the security rationale for keeping data briefly. EFS is Anthropic's attempt to close that gap: it pairs zero data retention with the abuse monitoring Anthropic says it needs to catch its models being used to hack companies or cause other harm. It also lands a month after OpenAI's own Private Safety Processing, suggesting that reconciling privacy commitments with safety monitoring has become a live point of competition among frontier model providers.

Who it affects

Enterprise customers of Fable 5.1 who use Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Claude Platform on AWS, Google's Agent Platform or Microsoft Foundry can apply for EFS by invitation once it ships. Mythos stays under Anthropic's earlier policy and is not covered. Consumer plans, Claude Free, Pro and Max on the web, desktop and mobile, plus Claude.ai and Claude Code, keep their previous terms and are unaffected.

How to use it

ZDR under EFS is not automatic: an enterprise must apply and be approved. Without such an agreement, the default for commercial API customers is 30 days of retention for inputs and outputs. Under EFS, the customer's own cloud infrastructure holds the data instead of Anthropic's, and Anthropic's automated monitoring sends flagged patterns straight to the customer to review, meaning the customer takes on part of the abuse-review work Anthropic staff would otherwise do. Anthropic plans to release EFS this fall; no more specific date has been given.

How solid is it

The account rests on Anthropic's own statements to The Register, including its description of "substantial evidence of attempted misuse of AI models" and its explanation of how EFS stores and routes data. The article carries no independent verification of these claims: no incident counts or examples of the misuse Anthropic describes, no named enterprise customers who complained about retention, and no pricing detail for EFS or ZDR agreements.

Risks and caveats

The article lists Claude Code both among the enterprise offerings eligible for EFS and among the consumer plans that keep the earlier retention policy, without explaining the overlap. Anthropic's assurance that the temporarily retained data is not used for training, like its abuse-detection claims, is not independently checked in the piece. Enterprises that take up EFS also take on the work of reviewing monitoring alerts themselves, a cost that offsets part of the compliance benefit.

“EFS works by storing data in cloud infrastructure controlled by the customer, not Anthropic”

— Anthropic