Anthropic Restores Global Access to Claude Fable 5 | U.S. Lifts Export Ban

SAN FRANCISCO | The standoff between one of the world's most highly valued artificial intelligence startups and the federal government is officially over.

Anthropic announced that it is restoring global access to its most advanced AI models, Claude Fable 5 and Claude Mythos 5, starting July 1, 2026. The rollout comes directly after the U.S. Department of Commerce lifted a sudden export control order that had forced the San Francisco-based company to pull both tools offline globally for over two weeks.

The reversal marks the end of a messy, unprecedented chapter in AI regulation. The Trump administration's initial decision to block the models highlighted a growing federal anxiety over the potential militarization of commercial AI and the lengths Washington is willing to go to maintain technological dominance. The incident sits within a broader pattern of the current administration asserting direct authority over the technology sector, visible also in Trump's sweeping crypto regulatory posture in 2025, and raises fundamental questions about how government security reviews will interact with the commercial AI release cycle going forward.

The Jailbreak That Triggered a Blackout | June 12 Commerce Department Action

The crisis began on June 12, just days after Anthropic launched the highly anticipated Fable 5. The Commerce Department abruptly issued an export control directive demanding that Anthropic restrict foreign nationals from accessing both Fable 5 and its enterprise-focused sibling, Mythos 5. Because Anthropic had no reliable mechanism to verify user nationality in real-time, the company opted to take both models completely offline worldwide rather than attempt a patchwork geographic restriction that could be trivially bypassed.

The trigger was a jailbreak. Amazon researchers reportedly discovered a prompt that successfully bypassed Fable 5's safety guardrails, tricking the model into identifying software vulnerabilities and, in one documented case, writing code that demonstrated how to exploit a specific flaw. Anthropic publicly argued the jailbreak was relatively unsophisticated and could be replicated against rival models including OpenAI's GPT-5.5 and China's Kimi K2.7. U.S. intelligence agencies and the Commerce Department disagreed sharply. Their concern was that the technique could theoretically arm foreign adversaries with automated cyber-attack capabilities at a scale and speed previously unavailable without state-level resources.

How Anthropic Won the Government Back | The 99% Classifier

Over the two-week period, Anthropic worked aggressively to train a new safety filter, a specialized classifier designed specifically to detect and block the identified jailbreak technique. According to the company, the new safeguard now stops the exploit in over 99% of attempts. That fix was subsequently tested and independently approved by the U.S. Center for AI Standards and Innovation (CAISI), a federal body under the National Institute of Standards and Technology responsible for evaluating AI safety claims.

In a letter clearing the company, Commerce Secretary Howard Lutnick confirmed the ban's removal. He noted that Anthropic agreed to "proactively detect and address security risks" and to inform the U.S. government of any malicious activity identified through its systems. This represented a significant concession. Anthropic, which has positioned itself as the safety-first AI lab since its 2021 founding by former OpenAI executives, was now formally committing to act as an intelligence-sharing partner with Washington in a relationship that has no clear precedent in the commercial software industry.

More broadly, Anthropic has committed to providing designated government partners with pre-release access to future frontier models before public launch, a structural shift toward tighter coordination between Silicon Valley and Washington. The arrangement mirrors the approach that the Salt Typhoon telecom breach response pushed for with private infrastructure operators: mandatory advance disclosure to federal agencies as a condition of continued operation.

Where You Can Access Fable 5 | Rollout Timeline and Credit Changes

Starting July 1, Claude Fable 5 is once again available to users via Claude.ai, the Claude Platform, Claude Code, and Claude Cowork. Pro, Max, and Team subscribers will have Fable 5 included within up to 50% of their weekly usage limits through July 7, 2026, after which access transitions to a credit-based system. Enterprise access via AWS Bedrock, Google Cloud Vertex AI, and Microsoft Azure AI Foundry is being re-enabled over the coming days on a rolling basis.

The situation for Mythos 5 is considerably more restricted. The enterprise model features fewer safety guardrails and is explicitly designed for high-level cybersecurity tasks including offensive security research and vulnerability analysis. According to VentureBeat's reporting on the redeployment, the Mythos 5 access list includes cleared defense contractors, national laboratories, and a small number of major enterprise cybersecurity firms with existing government contracts. General public access, and international access of any kind, has not been announced.

The broader implications of this incident extend well beyond Anthropic. The Commerce Department's willingness to invoke export control authority over a domestic AI company's commercial product sets a clear precedent: any model capable of generating working offensive security code is subject to the same federal oversight as physical dual-use technologies. The Snowflake breach that exposed 165 companies earlier in 2026 demonstrated how quickly a single security gap in a widely accessible platform cascades into systemic risk across hundreds of organizations. AI model jailbreaks capable of writing exploit code represent an analogous threat vector that regulators are now treating with identical severity.

Frequently Asked Questions