The United States has banned foreign users from accessing Anthropic's most advanced AI models, Bloomberg reported, an unusual move that highlights the Trump administration's current willingness to exert control over this critical industry. At the same time, it serves as a reminder to Silicon Valley that they are developing and using a technology that is not yet fully understood and whose impact is highly variable.

Anthropic

On Friday, the U.S. government took the unprecedented step of ordering Anthropic to ban all foreigners from using its most advanced AI platform. The U.S. government issued this order after discovering that the Fable 5 model released by Anthropic a few days ago could be "jailbroken." The so-called "jailbreaking" refers to bypassing the security protection mechanism set by the model.

The rapid introduction of the move sparked widespread concern. Anthropic called the move an "overreaction" and warned that taking such a regulatory approach could "result in all leading-edge model developers being unable to deploy any new models."

Anthropic executives are currently in discussions with officials in Washington about the specific security concerns raised by the latter, people familiar with the matter said. One of the key officials involved in the negotiations was U.S. Treasury Secretary Bessent. He has previously issued warnings to Wall Street about the risks posed by cutting-edge AI models, such as Anthropic's Mythos. In theory, Mythos can find and exploit vulnerabilities in software faster than humans.

Intervene in AI company operations

The U.S. government’s response is the most significant intervention by government officials in an AI company’s operations to date. The move comes as leading AI startups such as Anthropic seek to go public. Anthropic is valued at more than $900 billion. This unexpected control measure may set a precedent for major AI model developers such as OpenAI, Google and Meta.

Today, the U.S. government is willing to use extraordinary powers to force AI developers to deal with potential security threats in accordance with government requirements, although the Trump administration stated in an executive order issued in June this year that it will not force Silicon Valley companies to obey government orders.

“The United States’ cutting-edge models are increasingly viewed as strategic assets, and their use is strictly controlled and influenced by national security considerations,” said Gary Tan, portfolio manager at asset management company Allspring Global Investments. “This trend is likely to continue because the United States has the lead in computing power.”

Mythos model

Anthropic released Fable 5 last week, the first public version of its Mythos-class model. The company has repeatedly warned about Mythos' cyber capabilities, and White House officials have taken these concerns seriously, creating a path through voluntary executive orders for U.S. agencies to use Mythos and other cyber-capable models to patch vulnerabilities in their systems.

Washington's concerns may go beyond "jailbreaking" or the ability to affect a particular platform. The broader concern is that adversaries might try to steal an AI model’s weights, a highly valuable digital file that holds the key to a model’s capabilities. The decision to ban access to Fable and Mythos comes amid growing concerns in Washington about AI companies protecting trade secrets. The U.S. government believes that these companies need to guard against potential threats from insiders to prevent the leakage of core technologies.

After decades of spreading cutting-edge American technology around the world as a diplomatic and economic tool, Washington is now moving in the opposite direction. Now, the United States is increasingly inclined to keep technology for domestic use due to national security concerns.

"This shows that the United States is rapidly using export controls to curb the risks posed by cutting-edge AI." said Stefanie Kam, an assistant professor at the Institute of Defense and Strategic Studies at Nanyang Technological University in Singapore.

Silicon Valley boycott?

Policy goals aside, Silicon Valley is likely to resist Washington's attempts to tighten control. From Meta to OpenAI, major AI developers are vying to seize the commanding heights of this technology, because this technology is likely to reshape the entire economy and many industries.

However, it’s impossible to test every combination of an AI model’s potential capabilities, leading to vague concerns about threats that continue to come under scrutiny as the models continue to improve.

"There may be important capability thresholds. For example, AI systems that can automate large amounts of AI research and development work in a short period of time. Such systems may introduce entirely new risks." U.S. Senator Jim Banks wrote this month in a letter to Bessant, National Cyber ​​Director Sean Cairncross, and Defense Secretary Pete Hegseth. "Industry experts have also expressed serious concerns about the difficulty of bringing such systems under human control."