A US government letter invoking national security authority forced Anthropic to disable Fable 5 and Mythos 5 globally, citing an undisclosed jailbreak technique.
Fable 5 and Mythos 5 disappeared from Anthropic's platforms at 5:12 p.m. ET on Friday. No deprecation notice, no API transition period. Both models were cut globally and simultaneously, leaving users mid-session with error messages where model names had been.
Behind the shutdown was a US government letter invoking national security authority. ZDNet reported that the government instructed Anthropic to disable access for "any foreign national, whether inside or outside the United States, including foreign national Anthropic employees." Rather than build nationality verification into its systems, Anthropic disabled both models for all users, foreign and domestic alike.
The jailbreak letter
According to Anthropic's announcement, the letter stated that the government "believes it has become aware of a method of bypassing, or 'jailbreaking' Fable 5." The letter did not specify the underlying national security concern. Anthropic's own framing shifted during the disclosure: the company initially described the communication as a "directive," then walked that back, clarifying it arrived as a letter that "did not provide specific details of its national security concern."
What Anthropic reviewed was a demonstration of "a specific technique" that identified "a small number of previously known, minor vulnerabilities." That phrasing is conspicuously careful. If the vulnerabilities were already known and classified as minor, the open question is what the government's demonstration revealed that was new, and whether Anthropic's prior classification of those flaws as minor will survive outside scrutiny.
Developers and researchers felt the impact within minutes. In Claude Code sessions, selecting Fable 5 returned an error stating the model "may not exist or you may not have access to it." Other Claude models stayed operational, which suggests the government's concern is specific to Fable 5 rather than a general objection to Anthropic's work.
Regulatory territory with no map
A government order compelling an AI lab to kill specific deployed models over a jailbreak is, as far as current public record shows, without precedent in the US. Export controls on artificial intelligence have historically targeted chips, model weights, and compute thresholds. Using national security authority to mandate a nationality-based access cutoff during active deployment is a different instrument entirely, one that places large language models into roughly the same administrative category as ITAR-controlled hardware.
The artificial intelligence security industry has flagged exactly this kind of exposure. NewCore, which closed a $66 million seed round this week, is building identity infrastructure on the premise that autonomous AI agents create attack surfaces that existing enterprise systems were never designed to handle, SiliconAngle reported. A jailbreak that draws a national security response illustrates in blunt terms what that infrastructure gap can cost.
Fable 5 had already attracted controversy before Friday. Earlier reports documented that the model was throttling responses to artificial intelligence researchers at rates other users did not see, a story that ZDNet noted generated widespread backlash before the shutdown. The incident adds a second layer of scrutiny to a model that barely completed its launch cycle.
Competing models and the broader market
Elsewhere, Google's Gemini 3.5 Flash stumbled on third-party benchmarks the same week. Fresh Android Bench data placed it sixth overall with a score of 63.7, behind its predecessor Gemini 3.1 Pro Preview at 72.4 and well behind OpenAI's GPT 5.5 at 74, which topped the list. Gemini 3.5 Flash's average run cost reached $147.1, the highest on the board, Android Authority reported.
That result captures a recurring tension in the AI model market: capability claims made at launch rarely survive contact with independent benchmarks. Google's Flash branding has historically traded on speed and price; this release delivered neither advantage over older alternatives.
What the Fable 5 episode exposes is a communication gap between frontier labs and their regulators. A letter that lacks specific details, compelling a company to disable global model access rather than develop a selective compliance mechanism, suggests neither party has built procedures for this scenario. The governance frameworks around artificial intelligence have not kept pace with how quickly these models are deployed and depended upon.
Anthropologic has not said whether Fable 5 and Mythos 5 will return, or on what timeline.
---
Frequently asked questions
What are Fable 5 and Mythos 5?
Two Anthropic large language models pulled from all platforms on June 13, 2026, following a US government letter citing national security concerns. Other Claude models remain available and unaffected.
Why did the US government demand the shutdown?
The government cited a claimed jailbreak technique for Fable 5, stating it believed it had become aware of a method to bypass the model's safety guardrails. The letter to Anthropic did not provide specific details of the national security concern.
Will Fable 5 and Mythos 5 come back?
Anthropologic has not provided a timeline or confirmed a return. Restoration likely depends on whether the company can address the government's jailbreak concern to regulators' satisfaction.
Does this affect Claude API access?
Yes. Anthropic disabled both models globally rather than building nationality-verification systems. Developers who had integrated Fable 5 via the API or Claude Code lost access effective 5:12 p.m. ET on June 13.
