Anthropic aims to give business customers more control over how data generated from the use of its AI models is stored. Companies must still retain data for 30 days, but under the new plans, they can store it in their own cloud environment.
Reuters reported this based on a source familiar with the plans. Bloomberg had already partially reported the news earlier. Anthropic has been working on revising its retention policy for several months and has involved more than 100 customers from highly regulated sectors in the process.
Data remains with the customer
The most significant change is not the retention period, but the location of the data. Anthropic continues to require business customers to keep relevant data available for thirty days. However, they will now have the option to store this data on their own cloud infrastructure. It is no longer necessary to do so with Anthropic.
In doing so, Anthropic is revising the policy it announced in June for its more powerful AI models. At that time, the company stated that all traffic from business customers via Fable and Mythos, as well as future frontier models, must be retained for 30 days. That data is intended to help identify and counter new forms of cyberattacks that leverage AI technology.
Bloomberg notes that Anthropic had already anticipated resistance when it announced that policy. The company acknowledged in a recent report that mandatory data retention would likely not go over well with customers accustomed to a “zero retention” policy. Anthropic also saw a risk to its commercial position if competitors did not retain such data.
According to Bloomberg’s source, Anthropic had already begun implementing the new system before customers began criticizing the retention policy.
OpenAI takes a different approach
The change comes as OpenAI is developing its own solution to the same problem. With Private Safety Processing, the company aims to identify patterns of potential misuse across multiple interactions, without needing access to the underlying customer data. With Zero Data Retention, that data remains on infrastructure managed by the customer.
OpenAI is currently testing Private Safety Processing with an initial group of customers and plans to roll out the system further starting in September.