Anthropic blocks all public access to Claude Fable 5, Mythos 5 after US government order – what businesses should do

Last night, the US government issued an unprecedented export control directive ordering Anthropic to immediately stop all access to its top-tier Claude Fable 5 and Claude Mythos 5 models to foreign nationals, citing unspecified national security authorities.

In response, Anthropic blocked everyone public access to both models, globally — meaning that no user in the world has access to them at this time, even paying enterprise customers and Anthropic employees internally. This is a huge blow and twist after the public release of Fable/Mythos 5 just three days prior.

Current Fable 5/Mythos 5 sessions will fail and new requests will automatically be directed to older, less capable models such as Opus 4.8. Anthropic says in a blog post that "We believe this was a misunderstanding and are working to restore access as soon as possible." and apologizes to his customers.

The sudden regulatory intervention serves as a stark warning to the corporate sector: centralized, cloud-based frontier models exist at the absolute mercy of government oversight and vendor compliance.

Did Pliny the Liberator’s public jailbreak catalyze the USG’s extraordinary actions against Fable/Mythos 5?

The government’s sweeping action follows a viral Fable 5 jailbreak released publicly on X on June 10 by the prolific jailbreak "Pliny the Liberator," who claims to have successfully bypassed the model’s guardrails to extract functional instructions for cyber exploits, explosives, and chemical synthesis pathways, specifically noting "birch reduction method" for methamphetamine.

Pliny outlined a highly sophisticated multi-agent attack that used a combination of "Unicode, Homoglyphs, Cyrillic," long-context reference tracking and a technique for breaking malicious requests into harmless tokens outside of distribution. The attacker then uses an already jailbroken Opus model to reassemble the benign pieces back into actionable, limited results.

Anthropic did not specify whether it was the jailbreak that precipitated the government order, and in fact noted that the information provided by the US government regarding the particular jailbreak was poorly documented, writing: "To date, the government has only given us verbal evidence of a potential narrow, non-universal jailbreak, which essentially consists of asking the model to read a specific codebase and fix any software flaws. It is our understanding that a potential jailbreak has been shared with the government."

The company claims that the revealed capabilities are "widely available" in other public models, explicitly naming rival OpenAI’s GPT-5.5.

Additionally, Anthropic cautions that downloading a commercial model instead of a non-universal jailbreak sets a regulatory standard that could "essentially halting all new model deployments for all edge model providers".

The Pentagon precedent and the need to duplicate and diversify AI in the enterprise

This sudden interruption of Anthropic’s latest and greatest AI models will no doubt cause some consternation for organizations relying primarily on the Claude API – as they should, even though they still have access to other, less powerful Claude models.

As I warned earlier this year when the Pentagon suddenly blacklisted Anthropic, enterprises can no longer afford—from an operational reliability perspective—to run mission-critical workflows on any kind an AI model or even a provider. Placing all your AI "eggs" in one basket, so to speak, creates a single, ultimately fragile point of failure from which recovery or mitigation becomes extremely difficult.

Of course, in this case, Anthropic helpfully notes that "access to all other Anthropic models will not be affected." And while the Opus 4.8 or other Anthropic models may already be the organization’s preferred choice given their lower cost or be seen as acceptable fallbacks, the reality is that the US government order was narrowly targeted in this particular case – who’s to say the government won’t ask for a block of in the future all AI models/products/services of a given lab?

We had an indication that enterprise AI customers needed to diversify their vendors earlier this year. Recall that in March 2026, Secretary of Defense Pete Hegseth called the Anthropic a "supply chain risk" after the company refused to allow the military to use Claude for mass home surveillance and lethal autonomous weapons without safety restrictions.

The knock-on effect led to a broad ban on the use of Anthropic in defense supply chains, denying contractors access overnight.

The lesson of the Department of Defense’s fallout remains critically relevant today. Any organization that builds agent workflows or production applications tied solely to a single closed API provider risks immediate operational failure if that provider faces a ban, cyberattack, or export control directive.

As a technical leader in the enterprise, your primary goal, if not already achieved, must be urgent diversify your AI offering — whether it’s from other cloud-based AI models and providers, or AI models running on on-premise or virtual hardware controlled by the enterprise.

At this stage, the diversification of enterprise AI providers is undoubtedly imperative to ensure that you can continue to run AI workflows without interruption.

Implications for the Enterprise: Sovereign Setting vs. Capability Frontier

The community reaction to Fable 5’s takedown reflects the rapid shift of enterprise computing to hardware sovereignty.

AI founder Alex Finn reached out to X to mark Anthropic’s shutdown as a "waking up" urges developers to run local models of home GPUs to insulate themselves from regulatory volatility.

"No company or government will EVER be able to take away your native models," Finn writes, warning that government overuse will only escalate as models approach artificial general intelligence (AGI), the stated goal of OpenAI and some other AI firms, in which an AI model becomes capable of performing the most economically valuable work tasks now performed by humans.

Competitors are already capitalizing on this sentiment; Chinese open source AI vendor MiniMax was quick to highlight the open weights/open source availability of its new borderline M3 model, contrasting its decentralized availability against Claude’s centralized vulnerability. In other words: enterprises can download and run M3 on their own hardware now without ever having to worry about any government stepping in to prevent access.

This dynamic presents a complex trade-off for CIOs and IT leaders:

• The sovereign advantage: Running local models with open weights on sovereign hardware provides absolute control, ensures data privacy, and immunizes the enterprise against sudden government export controls, vendor policy changes, or API rate limits.

• The Frontier Sacrifice: Adopting a purely local strategy means sacrificing cutting-edge reasoning, agent capabilities, and massive context windows inherent in the latest closed-API frontier models that require multi-billion dollar centralized computing clusters to operate.

The most sustainable way forward is an active backup architecture. Enterprises must design their systems to be model agnostic. By building intelligent routing layers that can dynamically switch from a boundary model like Fable 5 to a fallback with open weights or a secondary provider’s API the moment an outage or regulatory ban occurs, businesses ensure their operations survive the volatile intersection of AI scaling and government oversight.

Technology

#Anthropic #blocks #public #access #Claude #Fable #Mythos #government #order #businesses