CNBC reports Anthropic will walk back a data retention policy that had concerned business customers. The company said Tuesday it will replace the contested approach with a new system called Enterprise Frontier Safeguards, after receiving what it described as extensive customer feedback. The policy Anthropic is now walking back was introduced in June alongside Claude Fable 5 and Mythos 5, which CNBC describes as two of the company’s most advanced artificial intelligence models. Under that policy, Anthropic required 30-day retention for all traffic on those models, saying the data was needed to help defend against misuse and “complex and novel” cyberattacks. Anthropic had said the retained data would not be used for any non-safety purpose, including model training. That assurance did not settle the issue for many enterprise customers, according to CNBC, which reported that users still expressed concern about the retention requirement. Kate Jensen, Anthropic’s head of Americas, told CNBC the company spent hundreds of hours working with customers on an alternative. The result is Enterprise Frontier Safeguards, which Anthropic says will let businesses control how their data is reviewed, stored and managed while still enabling automated safety monitoring. The key design change, as described by Anthropic to CNBC, is that some safety scanning can happen without Anthropic personnel reviewing the customer’s data. Jensen said the company can perform the scanning it needs while keeping the process on customers’ systems and data, so they do not have to compromise their own privacy posture. Anthropic said it will not charge for Enterprise Frontier Safeguards. CNBC reports the controls are intended to work whether customers access Anthropic’s technology directly or through a cloud provider, a relevant point for enterprises that buy models through existing cloud procurement and infrastructure channels. The move lands in the middle of a high-stakes enterprise push for Anthropic. CNBC notes that the company generates the vast majority of its revenue from enterprise sales and that the market is important as Anthropic prepares for what is widely expected to be a major initial public offering. The company is also competing with rivals including OpenAI, which CNBC says recently introduced its own data retention solution. Anthropic had previewed the safeguards last week in the context of its expanded Salesforce partnership, according to CNBC. CEO Dario Amodei said then that the system was meant to help keep Salesforce customer data private and prevent models from running out of control, while Anthropic put significant effort into permissions for the integration. The new safeguards are not broadly available immediately. CNBC reports that Enterprise Frontier Safeguards will roll out in phases, with Anthropic aiming for broader availability this fall. Anthropic also announced Claude Fable 5.1 and Mythos 5.1 on Tuesday, saying the updated models improve coding, knowledge work and research capabilities. Who benefits: Enterprise customers with sensitive workloads benefit if the safeguards work as described, because they gain more control over review and storage while retaining automated safety monitoring. Anthropic also benefits if the change supports broader enterprise deployment. Who's exposed: Anthropic remains exposed if phased rollout timing or implementation details do not satisfy customers that objected to the earlier policy. AI vendors may face buyer scrutiny as data retention becomes part of enterprise model selection.