Anthropic's Fable 5.1 and Mythos 5.1 models are now subject to sweeping US export controls that effectively bar foreign nationals from accessing the company's AI services. According to reports from Bloomberg, the Commerce Department issued an export-control directive in June 2026, citing national security authorities, requiring Anthropic to suspend all access to its Fable 5 and Mythos 5 models by any foreign national. This unprecedented action affects all foreign nationals, whether inside or outside the United States — including the company's own non-citizen employees. A US official confirmed that the Commerce Department issued the letter, which Anthropic reported did not specify the underlying national security concern, as reported by Bloomberg. The order represents a significant escalation in AI export controls, with press coverage characterizing it as unprecedented in barring foreign-national access to a deployed commercial AI model.
Anthropic released Claude Fable 5.1 on Tuesday, introducing significant improvements to the company's flagship AI model. According to reports from Anthropic, the model is now available across multiple platforms including Anthropic's own platform, Amazon Bedrock, Google Cloud, and Microsoft Foundry. The company positions Fable 5.1 and its restricted sibling Mythos 5.1 as the world's most advanced models for coding and knowledge work. The latest update introduces preserved thinking blocks to protect against AI distillation, with the API now verifying that thinking blocks are sent back with the same system prompt, tools, and messages that produced them.
The new model demonstrates substantial improvements in key performance areas. As reported by Anthropic, Fable 5.1 achieved 52.6% on Terminal-Bench-Science 0.1, more than double the previous version's 24.7%. In coding performance, Fable 5.1 scored 55.8% on Terminal-Bench 4.0, ahead of Opus 5 at 52.3% and OpenAI's GPT-5.6 Sol at 37.3%. Business workflow scores nearly doubled to 31.4% on AutomationBench. The company has extended knowledge access to June 2026, five months later than Fable 5. The performance improvements come alongside the new preserved thinking blocks feature, which ensures Claude's reasoning steps are protected during multi-turn conversations.
The preserved thinking blocks feature affects new API accounts created after August 31, 2026 12:00:00 AM UTC. According to Anthropic, this update applies to new Claude Platform organizations, Amazon Bedrock accounts, Google Cloud Vertex AI projects, and Microsoft Azure Foundry projects created on or after this date. The company is taking a phased approach to enforcement, starting with new accounts where they see the highest concentration of distillation-related abuse. Existing accounts won't be affected for Fable 5.1, giving developers time to make their integrations compatible with this update. Users can opt into "non-strict" mode to continue conversations without thinking blocks, though the API will indicate which blocks were dropped from the model's view.
Despite the performance improvements, Anthropic has maintained its pricing structure. According to reports from Anthropic, Fable 5.1 still costs $10 per million input tokens and $50 per million output. However, the company has implemented significant cost savings through reduced cache reads, which fall 75% to $0.25 per million. Anthropic estimates that typical workloads are 25% cheaper and complex agent tasks are up to 45% cheaper with the new model. The preserved thinking blocks feature may also reduce costs by allowing the API to reuse cached prompts more often, further improving efficiency.
Anthropic has provided timelines for both the model's availability and the copy protection measures. According to reports from Anthropic, Fable 5.1 retires no sooner than September 1, 2027. The company states that every account will face the copy protection check on future models, ensuring continued protection against unauthorized distillation activities. The preserved thinking blocks feature will apply to all users for future models, building on existing anti-distillation measures like enhanced distillation classifiers and restrictions on transferring sessions. This comprehensive approach aims to make distillation campaigns more difficult to execute while maintaining legitimate applications of the technology.
