Vahid Online in English

Automated AI translations of Vahid Online reporting. Exported from the Telegram Channel. Original content also available on X/Twitter.

← Back to latest feed

13 Jun 2026, 14:24

The export control order, citing national security concerns, has prohibited access for "foreign nationals," both inside and outside the United States.

Anthropic, the maker of the AI assistant "Claude," announced on Friday, June 22, that following an export control order from the U.S. government, it has cut off access to its two advanced models, "Fable 5" and "Mythos 5," "for all users."

This order was issued under "national security authorities" and prohibits access for any "foreign national," whether inside or outside U.S. territory, including Anthropic's own foreign employees. Anthropic stated that it received this order at 5:21 PM Eastern Time. Since the company cannot instantly separate foreign users from others, it has been forced to deactivate both models for all customers.

Access to other models from the company, including Opus, remains unchanged.

According to Axios and The Wall Street Journal, the letter of this order was sent by Howard Latnick, the U.S. Secretary of Commerce, to Dario Amodei, CEO of Anthropic, and was drafted in collaboration with the Bureau of Industry and Security of the Department of Commerce.

The letter's text did not specify detailed security concerns, but according to a government official, Washington became concerned about national security risks after another company claimed it had managed to bypass the safety mechanisms of Mythos ("jailbreak" it).

The same official added that the government had previously tried to convince Anthropic to delay the release of these models, and after failing, issued the "export control" letter.

Anthropic has opposed this decision but stated that it will comply.

According to the company, the disclosed method is a "limited jailbreak"; in practice, it means requesting the model to read a code and fix its errors; vulnerabilities that are minor and already known, which other public models, including OpenAI's GPT-5.5, can also find.

Anthropic emphasized that Fable 5 has been released with protective measures far stronger than any previous model and has been tested for thousands of hours in collaboration with the U.S. government, the UK’s AI Safety Institute, and independent groups before its release.

The company described the restriction of access for a commercial model available to hundreds of millions of users due to a limited vulnerability as "disproportionate" and warned that if such a standard is applied across the industry, it would effectively halt the release of any advanced model.

This order was issued just three days after the unveiling of Fable 5 and Mythos 5, models that Anthropic had introduced as its most powerful systems. Both are built on a technical foundation, but only Fable 5 was released to the public with strict limitations, particularly in the areas of cybersecurity and biology; Mythos 5 was made available without these restrictions and only to a select group of trusted partners, including cybersecurity and infrastructure companies.

These two are continuations of "Mythos Preview," a model that attracted the attention of Wall Street and government officials this spring with its advanced cybersecurity capabilities and was distributed among a limited group under a plan called "Glasswing."

However, the significance of this situation goes beyond a single model and a single company: this is the first time Washington has used the export control lever concerning an advanced commercial model. This situation shows that the fate of a model can be determined not by the decision of the creator but by a government order.

The "nationality-centric" framework of this restriction also directly targets non-American users and companies.

Anthropic states that it accepts the government's halt on unsafe releases but insists on a legal, transparent process based on technical realities. An action that, according to the company, does not align with such criteria.

Anthropic has called this situation a "misunderstanding" and stated that it will publish more technical details within the next 24 hours and is working to restore access, although it has not specified a timeline for that.
@VahidOOnLine

📡 @VahidOnline

Mapped locations (1)

Map data © OpenStreetMap contributors