Anthropic announced on September 1 a new system called Enterprise Frontier Safeguards, giving business customers more control over how their data is reviewed, stored and managed. The move follows pushback over a policy introduced in June alongside the launch of Claude Fable 5 and Mythos 5, which required 30-day retention of traffic on those models to help the company defend against misuse and complex, novel cyberattacks.
Under the new system, companies can run automated safety monitoring without Anthropic reviewing their data by hand, and businesses will be able to keep the required 30 days of data on their own cloud infrastructure rather than Anthropic's servers. Kate Jensen, Anthropic's head of Americas, said the company spent hundreds of hours working with customers to develop the alternative. The safeguards come at no additional charge and will roll out in phases, with broader availability targeted for this fall.
Enterprise customers are Anthropic's primary source of revenue, and the change comes as the company reportedly moves toward an eventual IPO. The shift signals how AI providers are being forced to balance safety monitoring against enterprise clients' data-privacy demands as frontier models handle increasingly sensitive corporate workloads.