Summary: Anthropic introduced Enterprise Frontier Safeguards so safety-monitoring activity is stored in the customer’s cloud account under customer keys, policies, and audit logging. Anthropic runs automated misuse detection and sends flags to the customer’s security team, with no Anthropic employee reviewing the data. The phased rollout begins later in the fall across Claude, Bedrock, Google, and Microsoft environments.
Frontier capability against data custody
On September 1, 2026, Anthropic introduced Enterprise Frontier Safeguards (EFS), its answer to a dilemma that has held back adoption of its strongest models in regulated industries. Mythos-class systems such as Claude Fable 5.1 mark a step change in intelligence and agentic behavior, and with that the stakes of misuse climb—fraud, sophisticated cyberattacks, destructive behavior carried out autonomously. Catching those patterns depends on watching traffic across sessions and accounts over time, which until now meant Anthropic would hold enterprise data for 30 days.
Training was never the point of that retention. Anthropic says it has never trained on enterprise data without explicit permission and never will. Even so, for many enterprises—banks, healthcare providers, and law firms among them—data sitting outside their control was by itself enough to create compliance friction. EFS comes out of designing around that constraint rather than asking customers to accept it.
What EFS changes
The shift in architecture is easy to state. The activity data that safety monitoring depends on is stored in the customer’s own cloud account—Amazon S3, Azure Blob Storage, or Google Cloud Storage—under the customer’s encryption keys, access policies, and audit logging. Automated systems at Anthropic read a rolling window of that traffic for signs of serious misuse: attempts to build offensive cyber or biological capabilities, say, or traces of stolen credentials. When a pattern warrants attention, the flag goes straight to the customer’s security team. No Anthropic employee reviews the data.
That division—detection run by Anthropic, custody kept by the customer—was shaped by more than 100 customers across financial services, healthcare, manufacturing, telecom, law, retail, and the public sector. The collaboration included the Analysis and Resilience Center for Systemic Risk (ARC), whose members include chief information security officers from Goldman Sachs, Morgan Stanley, Citi, Bank of America, and Wells Fargo. Leaders from Comcast, KPMG, Mastercard, Salesforce, Visa, and others also contributed input.
The concerns customers raised
Three worries surfaced repeatedly in those conversations. The first is expense: becoming yet another “trusted data vendor” obliges an enterprise to notify its own customers, redraw contracts, and satisfy internal storage and auditing rules. The second is that many regulated firms work under rules tightly governing who may see privileged legal material, non-public information, or drug-safety reports; their own staff are already trained and cleared for that review work, so whoever acts on a flag ought to be one of their own. The third is that automated monitoring helps, yet a human reviewer still adds value by confirming real misuse and clearing false positives. EFS answers all three by leaving data in customer-controlled infrastructure and routing flags to the customer’s team.
Phased rollout and supported platforms
Rollout happens in phases, starting later this fall. To smooth the transition, eligible customers receive zero data retention (ZDR) on Fable 5 and Fable 5.1 until EFS is ready. The solution will be supported on Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Claude Platform on AWS, Google’s Agent Platform, and Microsoft Foundry.
Implications for product builders
For teams building on frontier models in regulated spaces, EFS clears away a structural blocker. The pattern deserves study even if you never become an Anthropic customer: keep the detection layer apart from the data custody layer, and hand the review loop to the customer’s own security team. That is more than a compliance checkbox—it is what lets organizations put AI into parts of the business they previously could not touch. As one service partner quoted in the announcement phrased it, the safeguards “allow us to apply AI in parts of the business that we wouldn’t have been able to before.”
EFS is no universal answer. It asks customers to run their own storage and review workflows, which adds operational overhead. For enterprises that already have those capabilities, though, it turns a hard privacy tradeoff into an architecture decision.