Anthropic [Reuters=Yonhap News]

The U.S. government has eased restrictions on the use of Anthropic's top-tier artificial intelligence (AI) model, 'Mythos 5,' but limited this access exclusively to domestic U.S. companies.

This partial easing of regulations comes approximately two weeks after the Donald Trump administration announced export controls on Anthropic's AI models.

However, South Korean companies such as Samsung Electronics and SK Hynix are still expected to be restricted from using 'Mythos 5.'

According to Reuters and The Washington Post (WP) on the 26th (local time), the U.S. Department of Commerce notified Anthropic in a private letter that day that only specific companies within the United States are permitted to use Anthropic’s AI model, 'Claude Mythos 5' (Mythos 5).

In the letter obtained by the media, U.S. Secretary of Commerce Howard Lutnick stated, "Sufficient security measures have been put in place for certain trusted partners to use the 'Mythos 5' model."

In effect, the U.S. government has mandated that only select domestic companies and organizations deemed safe by the government may utilize 'Mythos 5.'

The letter also stated that the U.S. government reserves the right to change the list of approved entities at any time.

While the specific list of approved entities has not been confirmed, the WP reported, citing anonymous sources, that it is expected to include about 100 U.S. companies and organizations.

In a statement released following the media reports, Anthropic said, "Today, the government notified us that we can resume deploying our most powerful cybersecurity model, 'Mythos 5,' to various U.S. agencies that operate and defend critical national infrastructure."

This is interpreted to mean that 'Mythos 5' will be used exclusively by vetted U.S. companies and agencies, while restrictions for foreign companies remain in place.

Consequently, South Korean firms such as Samsung Electronics, SK Hynix, and SK Telecom—which were previously known to have early access to 'Mythos,' the predecessor to 'Mythos 5,' through collaborations with Anthropic—are expected to remain without access.

'Mythos 5' is a next-generation AI model specialized for cybersecurity released by Anthropic. The model boasts powerful capabilities in detecting and defending against security vulnerabilities. However, it has also been pointed out that it is a double-edged sword, as it presents challenges that existing cybersecurity systems may struggle to address.

For this reason, Anthropic launched 'Project Glasswing,' a global AI security collaboration project, and provided the pre-release 'Mythos' model to vetted companies and institutions worldwide. The intent was to proactively identify software vulnerabilities through these collaborations.

It was reported that South Korean companies, including Samsung Electronics, were granted access to 'Mythos' when they joined Project Glasswing on the 2nd.

However, after Anthropic officially launched 'Mythos 5' and 'Fable 5' based on the 'Mythos' architecture, the U.S. government announced export controls on these cutting-edge AI products, effectively blocking access for South Korean firms like Samsung Electronics.

U.S. government control measures targeting cutting-edge AI services are becoming visible at other companies beyond Anthropic as well.

OpenAI unveiled its next-generation AI model, 'GPT-5.6,' on the same day. However, rather than making it available to everyone as in the past, the company decided to provide it first only to select institutions designated by the U.S. government.

This has sparked criticism that the U.S. government is arbitrarily controlling access to state-of-the-art AI models.

Within the U.S. AI industry, some point out that the government's unprecedented and broad export controls on advanced AI models could inadvertently undermine America's own leadership in AI.

Meanwhile, according to a report by The New York Times (NYT) citing anonymous sources, Anthropic and the U.S. government are also discussing the possibility of restoring access to 'Fable 5,' which has generated significant consumer interest.

Unlike 'Mythos 5,' which is permitted only for computing infrastructure companies and a small number of institutional clients, 'Fable 5' was released with guardrails to allow usage by general consumers. Essentially, the 'Mythos' technology was provided as the product 'Mythos 5' to professional enterprises and agencies, and as 'Fable 5' to individuals.

Previously, on the 12th, the Trump administration issued an export control directive to completely block all foreign nationals from accessing 'Mythos 5' and 'Fable 5,' citing national security concerns.

As Anthropic lacked the functionality to verify the nationality of its users, it was forced to block all user access to the two models.

U.S. media outlets reported that employees working at the approved U.S. companies and institutions are now able to use these models, even if they are foreign nationals.