1 hour ago
796
On June 26, the U.S. Commerce Department approved Anthropic’s request to resume selling Claude Mythos 5 to a select group of U.S. customers, allowing Anthropic to have its previous export-control order partially reversed. Two weeks prior to this order, the company was compelled to remove its two most advanced AI products, Mythos 5 and Fable 5, from the market. Meanwhile, Fable 5, the consumer version based on the same underlying architecture as Mythos 5, continues to be offline with no definite date for its redeployment.
The partial reversal of the original export-control order has changed Anthropic’s competitive landscape at a crucial juncture in the company’s growth. Anthropic filed a confidential S-1 on June 1 with the SEC in preparation for a proposed IPO.
Each additional day that these flagship models are offline allows competitors to fill that void with their own offerings. For example, only one day after the original export ban was imposed, Chinese company Zhipu launched its open-weight GLM 5.2 model. According to a CNBC report referenced by Cryptopolitan, Zhipu’s token traffic rose faster than any other Chinese model tracked in 2026.
Why the US restricted Anthropic’s Mythos and Fable models
On June 12, the U.S. Commerce Department issued a directive to Anthropic to block all users outside the U.S. from accessing Anthropic’s models, including Anthropic’s own non-American employees. Anthropic was unable to block foreign users quickly enough; therefore, both models were disabled for all users.
The reasons given by the U.S. government for the block include cybersecurity concerns regarding methods of breaking into or bypassing Fable 5’s security features to find vulnerabilities in the software. Anthropic publicly disagreed with the government’s position, calling the findings narrow. The company argued that other models available for purchase, such as GPT-5.5, developed by OpenAI, could provide similar results in finding vulnerabilities without having to use jailbreak techniques.
In addition to the jailbreaking dispute between Anthropic and the U.S. government, there are broader cybersecurity implications.
During a congressional hearing, Senator Mark Warner of Virginia said NSA Director General Joshua Rudd told him Mythos could identify weaknesses across nearly all U.S. classified computer systems within hours. The NSA also conducted Project Glasswing, a covert effort with Anthropic and U.S. intelligence agencies to find and remediate critical software vulnerabilities, according to AP reporting and Cryptopolitan’s earlier coverage.
A U.S. official later clarified that identifying vulnerabilities within hours did not mean Mythos could exploit them in the same timeframe. Still, the demonstration gave the administration political cover to act.
What the Commerce Department now allows Mythos 5 to do
According to a letter from Commerce Secretary Howard Lutnick to Anthropic co-founder Tom Brown regarding Anthropic’s models, the company appears to have made “significant progress” toward mitigating the risks associated with both Mythos 5 and Fable 5. The letter is dated June 26 and was seen by both Reuters and The Verge.
In addition to granting access to the Mythos model without the need for an export license to some identified U.S. partners, Anthropic will also be able to provide access to non-U.S. employees of these registered U.S. partners and its own non-U.S. employees.
A source familiar with the directive told Reuters that there are currently more than 100 companies and institutions approved to use Mythos 5, including a number of Fortune 500 companies.
Anthropic spokesperson Danielle Ghiglieri confirmed the company had “received notice from the US government that Mythos 5, our strongest cybersecurity model, can be redeployed to a small group of cyber defenders and infrastructure providers,” according to The Verge.
Lutnick’s letter, however, did not reference the Fable 5 model. A source told Reuters that the U.S. government is progressing toward a resolution on the release of Fable 5 to the public, but no anticipated timeline has been established.
How the shutdown helped Chinese AI rivals gain ground
The ban on Anthropic’s Fable 5 and Mythos 5 was extremely inconvenient. The models were released on June 9 as Anthropic’s newest and best models, costing $10 per million input tokens and $50 per million output tokens. Three days later, both were forced offline.
Zhipu’s GLM 5.2 launched to subscribers on June 13, one day after the shutdown. This model, which has 744 billion parameters, has scored within four points of Anthropic’s Opus 4.8 on Terminal-Bench 2.1 and costs approximately one-fifth of the Anthropic model, Cryptopolitan reported previously. Full open weights of GLM 5.2 were shipped on June 16 under an MIT license, meaning that a government cannot retroactively deny access to anyone after downloading it.
OpenAI has a similar, but softer, limitation. The White House requested that OpenAI provide customers access to its latest model in an incremental format based on government approval from customer to customer.
Axios reported that GPT-5.6 launched under restricted access, with only around 20 government-approved companies getting limited preview access. This similar gating model will now also apply to providing access to Mythos 5. Two of the three largest U.S. frontier labs now operate with government gate requirements.
Why Anthropic’s government tensions matter before its IPO
The Mythos suspension was not an isolated event. Throughout much of 2026, Anthropic’s relationship with the U.S. government was strained. The company would not permit the U.S. military to use its models to conduct domestic surveillance or work with fully autonomous weapons; the government reacted by placing the company on its national security blacklist. President Trump later changed his mind, telling Axios on June 19 that he no longer considered Anthropic to be a national security risk, an event covered by Cryptopolitan at the time.
In addition, on June 10, Anthropic informed the Senate Banking Committee that operators of Alibaba carried out 28.8 million Claude transactions via about 25,000 fraudulent accounts between April and June 2026. Anthropic described this as the biggest attempt at extracting capabilities from Claude that it had ever seen, according to Cryptopolitan. Previous such distillation attempts have been made by DeepSeek, Moonshot, and MiniMax.
What happens next for Fable 5 and Anthropic’s rollout
Lutnick’s letter retained all other restrictions set forth in the June 12 letter but reserved the right to modify the terms should circumstances require. For Anthropic, the next step is ensuring that Fable 5 becomes available again before its proposed IPO.
With each day the model remains unavailable to the public, enterprise customers continue to run cost-benefit analysis on open-weight alternatives from China that no government restriction can easily take back.
Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.
English (US)