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

The US government last night issued an unprecedented export control order ordering Anthropic to immediately stop all access to its top-tier models Claude Fable 5 and Claude Mythos 5 to foreigners, citing unnamed national security authorities.

In response, Anthropic blocked everything public access to both models, worldwide — meaning no users worldwide can access them at this time, not even paying business customers and Anthropic employees internally. It’s a huge blow and turnaround following the public release of Fable/Mythos 5 just three days earlier.

Current sessions of Fable 5/Mythos 5 will end with errors and new quests will be automatically ported to older, less capable models such as Opus 4.8. Anthropic says so in a blog post "We believe this is a misunderstanding and are working to restore access as soon as possible," and apologize to its customers.

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

Did Pliny the Liberator’s public prison break trigger the USG’s unusual action against Fable/Mythos 5?

The intense government action follows the Fable 5 prison break that was made public on X on June 10 by a prison judge. "Pliny is an excellent host," that have successfully bypassed the model’s security lines to execute commands for cyber exploits, explosives, and chemical synthesis methods, especially "how to reduce birch" for methamphetamine.

Pliny described a more complex, multi-agent attack that used a combination "Unicode, homoglyphs, Cyrillic," long context reference tracking, and how to break malicious requests into innocent, non-distribution tokens. The attacker then used the previously jailbroken Opus model to separate the positive pieces and put them back together into possible, finite outcomes.

Anthropic doesn’t specify if this is the jailbreak that prompted the government order, and in fact, notes that the information provided by the US government about the particular jailbreak was poorly documented, writing: "So far, the government has given us only verbal evidence of a possible jailbreak that is small, not universal, which involves asking the model to read a specific codebase and fix any software errors. Our understanding is that one prison may be shared with the government."

This company is against open skills "widely available" in some public models, it clearly names OpenAI’s GPT-5.5 competitor.

In addition, Anthropic warns that pulling a commercial model over a non-universal prison sets a regulatory standard that could "completely stop the use of new models for all parameter model providers".

Pentagon precedent and the need for enterprise AI multiplicity and diversification

This sudden shutdown of Anthropic’s latest and greatest AI models will no doubt cause some confusion for organizations that rely primarily on the Claude API – rightfully so, even though they still have access to other less powerful Claude models.

As I warned earlier this year when the Pentagon suddenly blacklisted Anthropic, businesses will no longer be able to afford – from an operational reliability perspective – to run critical workflows. anywhere one AI model or even provider. Putting all your AI "eggs" in a single basket, so to speak, creates a single point, at the end of which it fades where recovery or reduction becomes more difficult.

Admittedly, in this case, Anthropic notes that it helps "access to all other Anthropic models will not be affected." And although Opus 4.8 or other Anthropic models may already be the favorites of organizations given their low costs, or are considered acceptable setbacks, the truth is, the US government’s order was narrowly targeted. in this case – who says the government will not, in the future, seek a bloc all AI lab models/products/services?

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

The resulting failure led to a massive ban on the use of Anthropic throughout the defense supply chain, depriving contractors of access overnight.

The lesson from the Department of Defense is still relevant today. Any enterprise-building agent workflow or production applications tied solely to a single closed API provider is at risk of immediate operational failure if that provider is subject to a law enforcement, cyberattack, or export control order.

As a business technology leader, your top goal if not already met should be urgently differentiate your AI offering – regardless of whether some AI models are based on the cloud and providers, or AI models running on local hardware controlled by the enterprise or virtual.

At this point, the diversity of the enterprise AI provider is undoubtedly important to ensure that you can continue to implement AI workflows without interruption.

Business implications: independent setup versus frontier capabilities

The public’s reaction to the Fable 5 drop shows the rapidly changing business reckoning in hardware governance.

AI founder Alex Finn took to X to flag Anthropic’s shutdown as "wake up call," urges developers to use local models on local GPUs to protect against control fluctuations.

"No company or government will EVER be able to take away your models," Finn writes, warning that government overreach will only increase as models inch closer to Artificial General Intelligence (AGI), the stated goal of OpenAI and other AI companies, where the AI ​​model can perform the most economically important tasks now performed by humans.

Competitors are already capitalizing on this sentiment; Chinese open AI provider MiniMax quickly highlighted the open weights/open source availability of its new, border-class M3 model, comparing its isolation availability to Claude’s medium vulnerability. In other words: businesses can download and run M3 on their hardware now without worrying about any government stepping in to block access.

These variables present a complex trade-off for CIOs and IT leaders:

• Sovereign Advantage: Implementing local, open-weight models on independent hardware provides complete control, ensures data privacy, and insures the business against government export controls, vendor policy changes, or API level restrictions.

• Frontier Sacrifice: Adopting a local strategy means sacrificing advanced thinking, agent capabilities, and large context windows associated with recent closed API models, which require centralized, multibillion-dollar computing clusters to run.

The strongest way forward is an effective way back. Enterprises must design their systems to be model-agnostic. By building intelligent routing layers that can dynamically switch from a frontier model like Fable 5 to open-weights fallback or a third-party API during outages or legal restrictions, businesses ensure their operations survive the dynamic intersection of AI scaling and government oversight.