According to CoinMeta, OpenAI has introduced a "zero data retention" policy as a response to Anthropic's 30-day data retention requirement. The new service Private Safety Processing previewed by OpenAI allows eligible companies and API customers to maintain zero data retention when using cutting-edge models. The system is capable of detecting potential abuses across multiple rounds of conversations, but OpenAI employees are not able to view customers' prompt or model responses. Customer data can continue to be stored on their own servers or in OpenAI, but it must be encrypted with the customer's own keys, and OpenAI cannot obtain these keys. Upon identifying risks, the system will only send limited security signals to OpenAI and will not transmit the original conversations. Meanwhile, Anthropic is also preparing to relax its 30-day data retention policy; corporate customers will still need to save 30 days of input and output data when using models such as Fable and Mythos, but they can keep this data in their own clouds without having to hand it over to Anthropic. This initiative has been in preparation for several months and is being jointly promoted by Anthropic and more than 100 customers, including Salesforce.