Identity Checks Mandatory for Claude Fable 5 After US Ban

Anthropic has restored developer access to Claude Fable 5 following a U.S. export control directive that forced the temporary global suspension of its newest AI models. The U.S. Department of Commerce ordered the shutdown on June 12, explicitly prohibiting access by foreign nationals after researchers demonstrated the model’s capacity to identify software vulnerabilities and generate exploits.

To comply with the order, Anthropic disabled both Fable 5 and the unrestricted Mythos 5 globally within three days of their launch, citing an inability to verify user citizenship in real time. The intervention was reportedly triggered by Amazon CEO Andy Jassy, who escalated security risks directly to the White House and Treasury Secretary Scott Bessent after internal testing defeated the model’s built-in safety filters.

Export Controls and Industry Reaction

The application of export controls to AI models represents a structural shift in U.S. policy, which previously focused on hardware restrictions. White House advisor David Sacks publicly stated the government forced the shutdown only after Anthropic initially declined to immediately de-deploy the models or patch the identified jailbreaks.

The cybersecurity community responded critically to the government mandate. Hundreds of researchers signed an open letter characterizing the export ban as misguided. Security experts noted that the identified jailbreak was effectively the model executing its intended defensive function of fixing software bugs, a capability already present in other broadly available systems.

Platform Controls and Verification

As of June 18, Fable 5 is back online with strict new compliance gates. Direct API access now requires formal identity and nationality verification for developers in specific jurisdictions.

For enterprise users, Anthropic is routing restored access through enhanced screening layers on major cloud providers. Developers can currently access the model via Amazon Bedrock, Microsoft Foundry, and Google Cloud Vertex AI. These platforms natively handle the identity verification and geographic restrictions mandated by the export control order. Legacy models, including Claude Opus 4.8, remain accessible without the new verification requirements.

Mythos-Class Specifications

Fable 5 and Mythos 5 represent a significant architectural leap over previous models, optimized for long-horizon agentic workflows and proactive self-verification. While the government order focused on risks surrounding multi-step cyberattacks, the underlying capabilities support state-of-the-art performance in general software engineering tasks.

Feature	Claude Fable 5	Claude Mythos 5
Target Audience	Public API users	Defense partners (Project Glasswing)
Context Window	1,000,000 tokens	1,000,000 tokens
Max Output	128,000 tokens	128,000 tokens
Input Cost	$10 per 1M tokens	$10 per 1M tokens
Output Cost	$50 per 1M tokens	$50 per 1M tokens
Security Guardrails	Conservative cyber classifiers	Unfiltered

The primary difference between the two deployments is the safety filter layer. Mythos 5 remains offline for the general public and is restricted exclusively to cleared partners through Project Glasswing for cyber defense applications.

Classifier Adjustments

The restored Fable 5 API includes aggressively tuned safety classifiers. Anthropic has adjusted the model’s refusal thresholds, specifically targeting prompts that ask the model to analyze or repair code. The Department of Commerce flagged “fix this code” requests as potential vectors for exploit generation, prompting Anthropic to implement highly conservative rejection logic for any prompt resembling vulnerability analysis.

If you depend on Fable 5 for automated code review or bug remediation, you must implement fallback routing to Claude Opus 4.8. The revised classifiers will actively reject standard debugging prompts that trigger the new exploit generation heuristics, making Fable 5 unreliable for direct code repair workloads.