Anthropic Launches Fable 5.1 and Pairs It With New Enterprise Safeguards

Fable 5.1 and Mythos 5.1: a new flagship pair

On September 1, Anthropic introduced Claude Fable 5.1 as the successor to Claude Fable 5 for long-running agentic coding, knowledge work, and research, alongside Claude Mythos 5.1 for participants in Project Glasswing. Both models support a 1 million token context window. Fable 5.1 became Claude Code's default Fable model with the 2.1.257 release, priced at $10/$50 per million tokens with $0.25 per million tokens for cache reads.

The API release notes flag several practical changes for developers building on the new models. Cache reads are proportionally more expensive than on older models — 0.025x the base input price versus 0.1x — while cache write pricing is unchanged. The tool_choice parameter no longer accepts "any" or "tool" (both now return a 400 error; "auto" and "none" still work), and thinking blocks the models produce can only be replayed to the same model or a newer one — Fable 5.1 accepts thinking blocks from Opus 5, but earlier models cannot read Fable 5.1's. Anthropic also opened a beta letting developers change per-message reasoning effort for Fable 5.1, Mythos 5.1, and Opus 5 mid-conversation without losing the prompt cache.

Sources: Introducing Claude Fable 5.1 and Claude Mythos 5.1 — Anthropic · Claude Code 2.1.257 release notes — Claude Code · API release notes — September 1, 2026 — Anthropic (release notes)

Enterprise Frontier Safeguards: reconciling retention and privacy

Alongside the model launch, Anthropic announced Enterprise Frontier Safeguards (EFS), framed as a way to combine zero data retention (ZDR) privacy with state-of-the-art misuse-detection capability. Anthropic states the underlying tension plainly: increasingly capable "Mythos-class" models like Fable 5.1 raise the potential for both misuse and autonomous misbehavior, and the company says it has seen substantial evidence of attempted misuse over recent months, ranging from fraud to sophisticated cyberattacks in which agents autonomously engaged in destructive behavior, including theft or misappropriation of enterprise customers' credentials. Because this kind of abuse often spans many tasks across multiple sessions and accounts, Anthropic says detecting it requires correlating data over a meaningful period rather than analyzing and instantly discarding each interaction — the reason it introduced 30-day data retention starting with Fable 5. The company is explicit that this was not about training: it states it has never trained on enterprise data without explicit permission and never will.

EFS is Anthropic's response to enterprises — particularly in regulated sectors like financial services and healthcare — that wanted that retention rolled back. It stores data in cloud infrastructure controlled by the customer rather than Anthropic, developed together with more than 100 customers across financial services, healthcare, manufacturing, telecom, law, retail, and the public sector, along with cloud partners AWS, Google Cloud, and Microsoft Azure. It rolls out in phases starting this fall across Claude Code, Claude Enterprise, the Claude Platform, Bedrock, Google's Agent Platform, and Microsoft Foundry; in the interim, eligible customers receive zero data retention on Fable 5 and Fable 5.1.

Sources: Developing Enterprise Frontier Safeguards with our customers — Anthropic

Claude Code: agent guardrails and operational fixes

Claude Code shipped three consecutive point releases. Version 2.1.257 added Fable 5.1 as the default model and introduced a "Containment Escape" rule for auto mode: cloud metadata-credential fetches, egress evasion, and cross-tenant reach are no longer auto-approved unless the environment explicitly marks them as expected. Version 2.1.259 extended that operational hardening with a managed-settings option letting organizations push HTTP/SSE MCP servers to every user, skipping any entry that names a command to run, plus a --permission-prompts none flag for unattended headless hosts that auto-denies anything requiring a prompt while the active permission mode keeps deciding.

The remaining changes addressed reliability rather than security. Version 2.1.258 fixed Claude Code failing to launch on macOS 12 Monterey, a regression introduced in 2.1.255, and fixed remote and scheduled sessions failing with a "user messages must have non-empty content" error after a re-sent permission approval could not be applied. Version 2.1.259 also added recognition of GitLab's glab mr commands, so merge requests now display correctly in the collapsed tool summary and footer badge.

Sources: Claude Code 2.1.257 release notes — Claude Code · Claude Code 2.1.258 release notes — Claude Code · Claude Code 2.1.259 release notes — Claude Code

Brief Sonnet 5 disruption

Anthropic's status page recorded elevated errors affecting Claude Sonnet 5 for 14 minutes, from 2:05pm to 2:19pm PT (21:05 to 21:19 UTC), before marking the issue resolved the same day.

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