Anthropic plans to revise a controversial data-retention policy for its most capable artificial intelligence models, according to Bloomberg, moving toward a setup that gives enterprise customers more control over where retained data is stored. The current plan, Bloomberg reports, is not to eliminate retention. Instead, Anthropic would still require enterprise customers to retain relevant data for 30 days, but would give them the option to keep it on their own cloud computing infrastructure rather than on Anthropic’s systems. The company expects to roll out the new safety system later this year, according to a person familiar with the matter cited by Bloomberg. Anthropic declined to comment to Bloomberg. The policy at issue dates to June, when Anthropic said it would retain all customer data for its more powerful Mythos and Fable models, along with future frontier offerings, for 30 days. The stated purpose was safety: Anthropic said the window would help it spot and prevent new forms of cyberattack using its technology. Bloomberg also reports that Anthropic said at the time it would not use the retained customer data to train its AI. The business tension was clear. Bloomberg reports that Anthropic acknowledged in a recent report that the policy would be unpopular with customers accustomed to zero retention and could create risks for its business if competitors did not adopt similar requirements. The Decoder also reports that Anthropic described the rule as unpopular and a business risk. The Decoder, citing Bloomberg, says Anthropic has spent months building the new system with more than 100 customers in regulated industries. Bloomberg similarly reports that the system has been in the works for months and that Anthropic has coordinated with more than 100 such customers. The Decoder adds that Anthropic developer Boris Cherny confirmed the plans publicly on X and says the changes are set to arrive this fall. One point needs careful handling: The Decoder’s headline frames the change as following enterprise pushback, while Bloomberg reports that Anthropic began work on the newer system before users began expressing criticism. The evidence supports saying Anthropic knew the retention rule created customer and business friction; it does not support a clean claim that criticism caused the policy change. Anthropic is not alone in trying to reconcile safety monitoring with enterprise data-control demands. Bloomberg reports that OpenAI said Wednesday it had started testing its own method, called private safety processing, with companies including Databricks and Microsoft, a major OpenAI investor. Bloomberg says OpenAI plans to release that approach more widely in September, while The Decoder describes it as a different method aimed at pairing security with data control. Who benefits: Regulated-industry customers stand to benefit if they can satisfy Anthropic’s safety requirement while keeping retained data in their own cloud environments. Anthropic may also benefit by reducing friction around use of its most capable models. Who's exposed: AI labs that impose retention rules without customer-controlled storage options may face more pressure from enterprise buyers. Anthropic remains exposed if customers view any mandatory 30-day retention period as incompatible with their data policies.