Anthropic reveals Enterprise Frontier Safeguards to monitor AI misuse without storing customer data

EFS will keep relevant activity data inside customer-controlled cloud infrastructure and will not store prompts and outputs, matching enterprise-grade protections. Automated monitoring sends any flagged issues to the customer, keeping Anthropic workers from handling potentially sensitive data. The company said it developed EFS with more than 100 customers in industries including financial services, healthcare, manufacturing, telecom, law, retail, and the public sector.

Anthropic is worried its Fable 5 and Fable 5.1 models could be used for cyberattacks, fraud and other autonomous threats. Detection of such abuse can require correlating behavior across multiple sessions, users and accounts, so the company earlier introduced a 30-day retention policy for some models. Regulated industries and enterprise customers objected to that policy, TechRadar reported. With EFS, monitored data remains at the customer side and any review is done by the customer's own staff. “What we heard from many customers, especially those in regulated industries, is that the person doing that review needs to be one of their own,” Anthropic said.

As an interim measure, eligible customers will receive zero data retention on Fable 5 and Fable 5.1 until EFS is available to them through a phased rollout and sign-up, TechRadar reported.