AI company Anthropic PBC has received a directive from the US government to suspend access of Fable 5 and Mythos 5, its most advanced programs, to foreign users. “PBC” stands for “Public Benefit Corporation.”
Only American citizens will be allowed to see or use the two flagship AI models. The order does not only refer to access abroad, it also includes blocking access to all foreigners regardless of where they are, even foreigners who are in the US. Anthropic employees who are not US citizens are also included in this ban.
The US government stated on Friday, 12 June, that Anthropic must abruptly disable Fable 5 and Mythos 5 for all its customers to ensure compliance.
According to Anthropic, the US government’s letter cited national security concerns. However, the directive did not provide any specific details.
Other AI models, such as Claude Chatbot, Claude Sonnet, Claude Haiku, Claude Opus, and earlier-generation Anthropic models, are not affected, Anthropic explained.
The following citation comes from an Anthropic press release:
“The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.”
“The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Anthropic models will not be affected.”
When an organization such as a company receives a directive, it is an order explaining what must be done. It is mandatory.
Anthropic has apologized to affected customers and says it is seeking the restoration of access to non-US citizens.
Jailbreaking and other concerns
The US government says that it has become aware of a method of jailbreaking Fable 5. Jailbreaking, in this context, means finding ways to trick a system into breaking its rules. Imagine you have a hotel with security guards who will not allow guests to enter certain rooms. A clever guest, however, finds a loophole and gets into those forbidden rooms anyway. That is similar to jailbreaking an AI.
Here is another example of jailbreaking. An AI may be programmed not to give dangerous instructions. However, a clever user rephrases a question which tricks the AI into providing the information it was supposed to refuse. Put simply, jailbreaking is the practice of tricking an AI program or system into ignoring its built-in restrictions or safety rules.
Anthropic’s response
Anthropic disagrees with the US government’s reasons for shutting down its two flagship models for foreigners. The company says that the vulnerabilities were previously known, and their findings showed that they represented only minor software weaknesses.
In fact, other models have similar vulnerabilities. After reviewing jailbreak evidence, Anthropic says that the US government’s concerns are overstated. Safeguards are robust and the risk of negative consequences is negligible.
The government’s current action is the result of a misunderstanding. Regulations should be supported by evidence, the company added.
Fable 5 underwent extensive red-team testing. US government agencies and UK safety organizations assisted in thousands of hours of testing. Today, Fable 5 boasts industry-leading safeguards. Anthropic says that it did not find any evidence of a universal jailbreak.
According to the company, the security measures that are present in Fable 5 exceed those of all its previous models.
The business impact
This is not only bad news for Anthropic’s business, but also its customers. Companies, hospitals, government departments, and other organizations across the world have abruptly lost access, which has resulted in serious workflow disruptions.
Many countries worldwide have already started to question the US’ reliability as a business and military partner after its unilateral imposition of tariffs and anti-NATO stances. If major companies in Europe, for example, are suffering further because of new US government AI directives, they will opt to do business elsewhere, such as in China, Japan, and South Korea.
Today, European lawmakers and captains of industry are talking about boosting domestic AI, given that America has made access a geopolitical issue.
