Anthropic is preparing to give business customers more control over their data when using its most advanced AI models, a notable reversal from a retention policy the company introduced earlier this year specifically to help guard against cyberattacks.
A new safety system expected to launch later this year will still require enterprise customers to retain data for 30 days, but will let them store that data on their own cloud infrastructure instead of handing it over to Anthropic, according to a person familiar with the plan who was not authorized to discuss it publicly. The system has reportedly been in development for months and grew out of coordination with more than 100 customers in highly regulated industries. Anthropic declined to comment.
The original policy, announced in June, called for Anthropic to retain all customer data generated by its most powerful models, including its Mythos and Fable systems and any future frontier models, for 30 days in order to detect and prevent new kinds of misuse of its technology. The company said at the time the data would not be used to train its AI. Even so, Anthropic acknowledged in a recent report that the approach would "be unpopular with customers who have come to expect zero retention, and pose real risks to our business success (especially if competitors do not follow)."
