Claude Fable 5 and Mythos 5 Suspended: The Full Story Behind the Shutdown

To understand the Fable 5 situation, it helps to start with Mythos. On April 7, 2026, Anthropic announced Claude Mythos Preview, a model that immediately stood apart from anything the company had previously released. It was not positioned as a general-purpose assistant upgrade. It was built with a specific and unusual purpose: finding software vulnerabilities at scale, across the most widely used operating systems and browsers in the world.

Anthropic chose not to release Mythos Preview to the public at launch, citing safety concerns around its cybersecurity capabilities. Instead, the company launched Project Glasswing, a controlled program that gave access to a small group of vetted partners including Amazon Web Services, Microsoft, Apple, Google, Cisco, CrowdStrike, NVIDIA, and JPMorganChase. These partners used Mythos Preview to identify and fix critical vulnerabilities in systemically important software. By late May 2026, Anthropic reported that the program had uncovered more than ten thousand high or critical severity vulnerabilities across major platforms.

Claude Fable 5 launched on June 9, 2026, as the public-facing version of the Mythos model family. Anthropic described it as a Mythos-class model made safe for general use, meaning it shared the same underlying architecture as Mythos but came equipped with additional safety classifiers and guardrails designed to prevent misuse in areas like cybersecurity, biology, and chemistry.

In practical terms, Fable 5 was Anthropic's most capable widely released model to date. It arrived with a one-million-token context window, adaptive thinking enabled by default, and performance that exceeded all previous Claude models across benchmarks covering software engineering, scientific research, knowledge work, and vision tasks. The model was available from day one across the Claude API, AWS, and Microsoft Foundry. It was designed for long-horizon agentic work where tasks are complex, extended, and require sustained reasoning across large amounts of information.

At 5:21 PM Eastern Time on June 12, 2026, Anthropic received an export control directive from the US Commerce Department. The directive, issued under national security authorities, ordered the company to suspend all access to Claude Fable 5 and Claude Mythos 5 for any foreign national, whether inside or outside the United States. The order explicitly included foreign national employees of Anthropic itself.

Because Anthropic cannot verify user residency in real time at the API level, the company made the decision to disable both models for every customer worldwide. The API string claude-fable-5 began returning 404 errors by approximately 9:59 PM Eastern Time that evening. Developers and enterprises who had already integrated Fable 5 into production workflows found their systems broken within hours of the shutdown. Claude Opus 4.8, Sonnet 4.6, and Haiku 4.5 were not affected by the directive and remained fully available.

The US government cited national security concerns and, according to Anthropic's own statement, communicated verbally that a method existed to jailbreak Fable 5 into bypassing its safety classifiers. The specific concern, as Anthropic described it, essentially involved asking the model to read a codebase and identify software flaws, a code analysis capability that security professionals use daily in defensive work.

Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei notifying the company that Fable 5 and Mythos 5 would be subject to export controls. According to reporting by Axios, the directive came after another company claimed it had found a way to jailbreak Mythos, which alarmed the administration. Under the directive, a license would be required for the export, re-export, or domestic transfer of both models to any foreign national.

Anthropic complied with the directive but made its disagreement publicly clear. In an official statement, the company said it acknowledged the government's authority to block model deployment but argued the process should be transparent and grounded in technical facts, and that this directive did not meet that standard. Anthropic described its understanding of the jailbreak as narrow and non-universal, and noted that the capability in question was already widely available in other commercial models including OpenAI's GPT-5.5.

Anthropic also raised a broader concern: if a narrow potential jailbreak is sufficient grounds to recall a commercially deployed model, the same standard applied consistently across the AI industry would effectively halt the release of any new frontier model. The company stated it was working to restore access as soon as possible but did not commit to a specific timeline. Anthropic also noted that the directive came against the backdrop of an already strained relationship with the US government, including a dispute earlier in 2026 in which the Department of Defense classified Anthropic as a supply chain risk, a classification the company has said it intends to challenge legally.

Claude Mythos 5, announced the same day as Fable 5 on June 9, 2026, is the latest iteration of the Mythos model and shows further gains in cybersecurity, biology, and healthcare benchmarks compared to Mythos Preview. Unlike Fable 5, Mythos 5 was never made publicly available. It remained restricted to Project Glasswing partners and a small group of vetted biology researchers.

Despite never having broad public access, Mythos 5 was included in the government directive alongside Fable 5. Both share the same underlying model architecture, which likely contributed to the decision to include both in the same export control action. The suspension of Mythos 5 affected Project Glasswing partners who had been using the model for defensive security research.

For any developer or business that integrated Claude Fable 5 into production systems during the three days it was available, the shutdown created immediate disruption. API calls started returning errors, automated workflows broke, and teams had to scramble to switch to a fallback model. Anthropic's recommended path for developers in this situation is to fall back to claude-opus-4-8, which is the closest available replacement in terms of reasoning and capability.

For businesses evaluating AI strategy more broadly, the incident highlights a risk that often goes unconsidered: dependency on a single model from a single provider exposes operations to disruptions that have nothing to do with technical reliability. Government directives, legal disputes, and safety decisions can all affect access to a model overnight. Building workflows that can fall back gracefully to alternative models or alternative providers is no longer a theoretical best practice. The Fable 5 situation made it a practical requirement.

Claude Opus 4.8 is the most capable model currently available from Anthropic and the most direct replacement for Fable 5 for demanding reasoning, coding, and agentic work. Claude Sonnet 4.6 remains an excellent choice for tasks where a balance between speed and capability is more important than maximum performance. Both models are fully available through the Claude API, claude.ai, AWS, and Microsoft Foundry.

For users whose workflows specifically relied on Fable 5's one-million-token context window or adaptive thinking features, Opus 4.8 handles large context tasks well and covers most of the same use cases. Teams building production systems during the suspension period are advised to set model selection at the configuration level rather than hardcoding a specific model string, so that switching models in response to future availability changes does not require code changes across an entire codebase.

As of June 17, 2026, Anthropic has not announced a timeline for restoring access to Claude Fable 5 or Claude Mythos 5. The company has said it is working to restore access as soon as possible and is in ongoing discussions with the government. Resolution likely depends on one of two paths: either an independent technical review of the alleged jailbreak determines that the capability does not meet the threshold for an export control action, or Anthropic deploys a targeted patch that addresses the specific concern raised.

If the dispute extends into legal or legislative territory, the timeline could stretch significantly. For planning purposes, it is safest to treat the suspension as indefinite and build workflows around currently available models. Anthropic has committed to communicating updates ahead of time so users and developers know where things stand before any changes take effect.