Anthropic has decided to change a controversial data retention policy following pushback from corporate customers. CNBC reported on Sept. 1 that Anthropic will roll out a new protection framework for business customers, called Enterprise Frontier Safeguards, in phases.

The policy at issue was introduced in June alongside the launch of Claude Fable 5 and Mythos 5. At the time, Anthropic said it would retain all traffic for those models for 30 days to respond to misuse and complex, new cyberattacks. Anthropic said it would not use the data for purposes unrelated to safety, including model training, but corporate customers raised concerns.

Under the new protection framework, businesses can directly control how data is reviewed, stored and managed. Automated safety monitoring is also possible without review by Anthropic staff. The feature will be offered at no extra charge and will apply both to direct use of Anthropic technology and use via cloud providers.

Kate Jensen (케이트 젠슨), Anthropic's head of the Americas, said the company consulted with customers for a long time to come up with an alternative. She explained that the system was designed to maintain scanning at the necessary level while processing it within customer systems and data so it does not undermine their own privacy standards.

The existing June policy will remain in place for non-corporate subscribers using Mythos-class models. Anthropic's new protection framework will be rolled out in phases.