Anthropic AI Updates: September 2, 2026

1. Anthropic Releases Claude Fable 5.1 and Claude Mythos 5.1

Anthropic. Fable 5.1 is the new frontier model for coding and knowledge work, scoring 73.4% on CursorBench 3.2.0, roughly doubling its predecessor on Terminal-Bench-Science, and improving agentic coding by more than 30%. Pricing is $10 per million input tokens and $50 per million output tokens, with cache reads cut 75% to $0.25 per million, which Anthropic says lands at about 25% cheaper for typical workloads and up to 45% cheaper for tool-heavy agentic runs. Mythos 5.1 is the same underlying model with cyber and bio capabilities retained, restricted to verified US organizations by invitation, and both models are available via API as claude-fable-5-1 on AWS, Google Cloud, and Microsoft Azure. Source

2. Enterprise Frontier Safeguards Puts Customer Data in the Customer’s Cloud

Anthropic. Enterprise Frontier Safeguards combines zero data retention with automated misuse detection by storing customer data in infrastructure the enterprise controls on AWS, Google Cloud, or Azure. The three opt-in components are customer-controlled cloud storage, customer-managed encryption keys, and fully automated safety monitoring with no Anthropic human review. Anthropic developed it with more than 100 enterprise customers across financial services, healthcare, manufacturing, telecom, law, retail, and the public sector, with rollout starting in fall 2026. Source

3. Claude Watermark Detection API Opens to Regulators and Researchers

Anthropic. Anthropic opened access to an API that checks whether text carries Claude’s invisible watermark, with approved access for regulators, law enforcement, media, fact-checkers, independent researchers, educational organizations, and EU civil society groups, plus enterprises verifying their own compliance. The watermark builds on Google DeepMind’s SynthID approach, biasing word-selection randomness into a statistically detectable pattern that survives some editing and carries no user data. The EU AI Act has required marking of AI-generated text since August 2025, and detection remains unreliable on short text, code, and heavily paraphrased output. Source

4. Claude Code v2.1.257 Adds Fable 5.1 and a Containment Escape Rule for Auto Mode

Anthropic. Claude Code v2.1.257 makes claude-fable-5-1 the default Fable model with a 1M context window. Auto mode gains a Containment Escape rule so cloud metadata-credential fetches, egress evasion, and cross-tenant reach are no longer auto-approved unless the environment marks them expected, plus a one-time prompt before the first file read outside the working directories and a permissions.blockReadsOutsideWorkingDirectories setting to block them entirely. Other additions include timeFormat and timeZone settings, CLAUDE_CODE_SUBAGENT_MODEL_FORCE to apply one model to every subagent, a session-only s option in /effort, and a /doctor warning for stale sandbox mask files. Source