Anthropic Eases Data Retention Rule After Backlash

What happened

Anthropic announced this week that it is revising a controversial data retention policy for its business customers after what the company described as extensive feedback since the rule took effect in June. The company is rolling out a new system called Enterprise Frontier Safeguards, which it says will give businesses greater control over how their data is reviewed, stored and managed.

The original policy, introduced alongside the launch of Anthropic's Fable 5 and Mythos 5 models, required 30 days of retention for all traffic sent to those systems, a measure the company said was necessary to help detect misuse and novel cyberattacks. Anthropic had pledged not to use the retained data for training or other non-safety purposes, but many enterprise customers pushed back regardless.

Why it matters

The reversal highlights the tension AI companies face between building in strong safety monitoring for increasingly capable models and reassuring enterprise customers, particularly those in regulated industries, that their sensitive data remains under their own control.

Anthropic generates the bulk of its revenue from enterprise customers, making the market's reaction to the original policy a meaningful business risk, one the company acknowledged directly in explaining its decision to change course.

Evidence and context

Under the new Enterprise Frontier Safeguards system, the required 30-day safety-monitoring window remains in place, but the underlying data will be stored within the customer's own cloud environment rather than on Anthropic's servers. Businesses will also be able to conduct automated safety monitoring without requiring direct human review by Anthropic staff.

Anthropic said it spent hundreds of hours working with more than 100 enterprise customers, including companies in regulated industries, to develop the new approach, which is expected to roll out in phases with broader availability targeted for this fall. The company said it will not charge customers for the new safeguards, and the controls will apply whether businesses access Anthropic's technology directly or through a cloud provider.

We can do the scanning that we really need to, to help everybody feel confident, but it's on their systems and their data.

— Kate Jensen, Anthropic's head of Americas

The people and places affected

Enterprise customers across regulated sectors were among the most vocal critics of the original policy, with reports indicating that at least one major technology company restricted its own employees' access to Anthropic's Fable 5 model over data retention concerns before this change was announced.

The June policy continues to apply unchanged to non-enterprise subscribers using Anthropic's Mythos-class models, meaning the rollback is specifically targeted at the business customer segment that raised the most significant objections.

The open questions

It remains to be seen how smoothly Anthropic can implement the technically complex shift of moving safety-monitoring data into each customer's own cloud infrastructure across its full enterprise customer base by the fall target.

The episode also raises a broader question for the AI industry about how companies balance safety monitoring capabilities against enterprise data control demands, a tension that rival OpenAI is separately navigating through its own alternative approach built with Databricks and Microsoft.

What comes next

Anthropic plans to roll out Enterprise Frontier Safeguards in phases over the coming months, with the company targeting broader availability this fall as it works through the technical implementation with its enterprise customer base.

The shift will likely be watched closely by rival AI labs as a potential template for balancing safety requirements with enterprise data control demands, particularly as more companies adopt AI tools that require handling sensitive corporate information.