30-day retention is out for Anthropic’s enterprise users

Anthropic said on September 1 that it will drop mandatory 30-day retention for business customers on Claude Fable 5 and Mythos 5 and replace it with Enterprise Frontier Safeguards, a free set of controls that keeps the required safety scanning on the customer’s own systems, CNBC reported. Enterprise teams get to keep their traffic instead of handing Anthropic a month of it, but the system rolls out in phases, with broader availability targeted for this fall. Non-enterprise subscribers on Mythos-class models stay under the June rule.

What the June rule required

Anthropic introduced the retention policy in June, alongside the launch of Fable 5 and Mythos 5. It required 30 days of retention for all traffic on those models, which the company said would help defend against misuse and “complex and novel” cyberattacks. The retention covered prompts and completions on Anthropic’s own surfaces and on third-party platforms, and it overrode existing zero data retention agreements, the contractual promise that a provider stores none of the prompts or outputs sent through its API.

Anthropic pledged not to use the retained data for any “non-safety-related purpose,” model training included. Many enterprise users were still concerned.

What Enterprise Frontier Safeguards changes

Under the new system, businesses control how their data is reviewed, stored and managed, and they can run automated safety monitoring that requires no human review by Anthropic. The controls work whether a customer reaches Anthropic’s models directly or through a cloud provider. Anthropic says it will not charge for the system.

The monitoring runs on the customer’s own systems and data, letting businesses carry out the mandatory safety testing without giving Anthropic access to the underlying data, Mint reported.

Kate Jensen, Anthropic’s head of Americas, said the company spent hundreds of hours with customers building the replacement. “We can do the scanning that we really need to, to help everybody feel confident, but it’s on their systems and their data so that they don’t have to compromise on their own privacy posture,” she said. Jensen added that she is “very confident” deployment will accelerate after the announcement, especially on customers’ most sensitive workloads.

The same day, Anthropic released Claude Fable 5.1 and Mythos 5.1, which it says improve coding, knowledge work and research.

Where OpenAI stands

Enterprises bring in the vast majority of Anthropic’s revenue, and CNBC frames the change as part of a race to stay ahead of rivals such as OpenAI. OpenAI said on August 19 that business customers can keep zero data retention and still get safety monitoring on its most advanced models through Private Safety Processing, which sends a narrowly defined safety signal without exposing prompts or responses. In August that system was in testing with early customers. Mint ties Anthropic’s change to OpenAI’s new data retention system, which it says will help OpenAI attract customers facing tight privacy and security regulations.

Dario Amodei previewed Enterprise Frontier Safeguards the week before, in a CNBC interview about Anthropic’s expanded partnership with Salesforce. He said the system helps keep Salesforce’s customer data private while making sure Anthropic’s models “don’t run out of control.”