Enterprise Frontier Safeguards: Your Data, Anthropic's Detection
On September 1, 2026, Anthropic launched Enterprise Frontier Safeguards (EFS) — an architecture that gives regulated enterprises zero-data-retention privacy and cross-session misuse detection at the same time. If compliance or security teams have blocked Claude adoption at your company, this changes the conversation.
The problem EFS solves: Anthropic introduced 30-day data retention with Fable 5 because sophisticated misuse — cyberattacks, credential theft, biological capability probing — unfolds across many sessions and accounts. Analyzing each conversation in isolation and instantly discarding it misses those patterns. But 30-day retention made frontier models unusable in regulated industries: banking, healthcare, law, and others where strict data-handling rules apply.
Enterprise Frontier Safeguards (EFS) moves where data lives. Instead of Anthropic storing your conversations, your activity data stays in your cloud account — Amazon S3, Azure Blob Storage, or Google Cloud Storage — under your own encryption keys and access policies. Anthropic's automated monitoring analyzes the data there. No Anthropic employee reviews your conversations.
Three opt-in controls — none change model behavior, API pricing, or rate limits:
Customer-owned storage: Conversation logs stay in your cloud account, not Anthropic's infrastructure.
Customer-Managed Encryption Keys (CMEK): You hold the keys; you govern who has access.
Fully automated review: Automated systems scan a rolling window of traffic for serious misuse signals and send flags directly to your team. No Anthropic human review is required.
Anthropic does not charge for EFS. If you opt into customer-owned storage, your cloud provider bills you for storage, reads, writes, and egress — same as any other resource in your account.
Built with 100+ enterprises across regulated industries. Anthropic developed EFS with customers in financial services, healthcare, manufacturing, telecom, law, retail, and the public sector — including a quarter of the Fortune 100 and every US global systemically important bank. The Analysis and Resilience Center for Systemic Risk (ARC), whose members include the CISOs of Goldman Sachs, Morgan Stanley, Citi, Bank of America, and Wells Fargo, participated in shaping the architecture.
EFS is supported on Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Claude Platform on AWS, Google's Agent Platform, and Microsoft Foundry. It rolls out in phases starting later in fall 2026. During the transition, eligible customers receive zero data retention on Fable 5 and Fable 5.1.
What this means for your Claude usage: If compliance or security teams have blocked AI adoption at your company because of data retention, EFS is the architectural answer to bring to that conversation. The split is concrete: your data stays in your environment, under your keys — Anthropic's automated systems analyze it for misuse signals, and those flags go to your team, not Anthropic's employees.
If you are building on Claude and your customers need institutional-grade data guarantees, EFS lets you offer frontier-model capability without a data-custody tradeoff. The phrase to carry into any internal conversation: data never leaves the customer's environment.
What you can do now that you couldn't before:
Use Claude in regulated workloads — financial, legal, healthcare — where 30-day retention was an adoption blocker.
Point your compliance and security teams to a concrete architectural answer: your cloud, your keys, your team reviews flags, no Anthropic human access required.
Use Claude to map your organization's specific regulatory requirements to EFS controls before bringing the conversation to legal or leadership.
EFS access is available by request form. Eligible customers receive zero data retention on Fable 5 and Fable 5.1 while the phased rollout completes later in fall 2026.