Claude's Enterprise Safety Net Moves Off Anthropic's Own Servers

Enterprise Frontier Safeguards: detection without Anthropic holding the data

Anthropic announced Enterprise Frontier Safeguards (EFS), a system that pairs zero data retention (ZDR) with misuse-detection capabilities by storing the relevant data inside cloud infrastructure that the customer controls, rather than Anthropic's own. The company built EFS with more than 100 customers across financial services, healthcare, manufacturing, telecom, law, retail and the public sector, alongside cloud partners AWS, Google Cloud and Microsoft Azure, and it will reach Claude Code, Claude Enterprise, the Claude Platform, Bedrock, Google's Agent Platform and Microsoft Foundry in phases starting this fall; eligible customers get ZDR on Fable 5 and Fable 5.1 in the meantime.

Anthropic's own account of why this was needed matters here: it says it introduced 30-day retention starting with Fable 5 because effective misuse detection — covering fraud, sophisticated cyberattacks, and cases of credential theft — often spans many tasks across multiple sessions and accounts, so scanning each interaction in isolation and discarding it instantly is not enough. The company is explicit that this was not about training on enterprise data, stating it has never done so without explicit permission and never will; it adds that customers generally understood the security rationale but many did not want Anthropic to be the one holding the data, which is the gap EFS is designed to close.

System card and research: what's behind the added caution

Alongside EFS, Anthropic published the system card for Claude Fable 5.1 and Claude Mythos 5.1 — the 'Mythos-class' models the EFS announcement itself cites as representing a major jump in intelligence and agentic capability, and therefore in the potential for both misuse and autonomous misbehavior. Read together, the two releases present EFS less as a standalone product move and more as the operational answer to risks the system card documents directly.

Anthropic also published a piece on formalizing Fermat's Last Theorem in Lean, a research exercise in machine-checked mathematical proof, and a guide to building Claude Commerce Agents aimed at applied, transaction-handling use cases. The pairing illustrates Anthropic's usual split focus: deep formal-verification research on one side, concrete enterprise agent tooling on the other.

Sources: Claude Fable 5.1 & Claude Mythos 5.1 System Card — Anthropic · Formalizing Fermat's Last Theorem in Lean — Anthropic · Building Claude Commerce Agents — Anthropic

Claude Code: three releases of governance and productivity tooling

Claude Code moved through 2.1.259, 2.1.260 and 2.1.261 in quick succession, with a noticeable emphasis on organizational control: managed HTTP/SSE MCP servers that admins can push to every user, a new '--permission-prompts none' flag that auto-denies anything requiring a prompt on unattended headless hosts, and a diagnostic line in /status and claude doctor explaining why an organization policy failed to load, such as a proxy not passing the endpoint through.

The same releases added day-to-day productivity features: a /diff panel showing uncommitted changes beside the conversation in fullscreen, higher inline output limits for commands and background tasks (up to 128K characters via bashOutputMaxChars and taskOutputMaxChars), visibility into prompt-cache miss causes in /cost, and recognition of GitLab merge-request commands. Separately, the Anthropic API's ant CLI reached version 1.30.0 with 'ant apply', which creates and updates agents, environments, skills, memory stores and deployments from files in a repository, tracked through a committed claude-lock.json lockfile — extending infrastructure-as-code practice to agent configuration.

Sources: Claude Code 2.1.261 — Claude Code · Claude Code 2.1.260 — Claude Code · Claude Code 2.1.259 — Claude Code · Anthropic API release notes — September 3, 2026 — Anthropic

Brief, resolved errors on the status page

Anthropic's status page recorded two separate incidents of elevated errors on September 6, both since resolved. One affected Claude Mythos 5.1, Claude Fable 5.1 and Claude Opus 5 together, where Anthropic said it had identified the cause and was working on a fix without detailing what that cause was; the other affected Claude Sonnet 5 alone and was logged only as under investigation before resolution. Neither incident report specifies duration or user impact beyond the elevated error rate itself.

Sources: Claude — Elevated errors for multiple models — status.claude.com · Claude — Elevated errors for Claude Sonnet 5 — status.claude.com