US bans foreign access to new Anthropic AIs: ‘National security’

The United States banned, this Friday, the 12th, foreigners’ access to Anthropic’s new artificial intelligence models; the country claims a “national security” risk. The measure is valid for foreigners inside and outside the USA, including startup employees.

The company had to take down the models globally for the directive to be complied with. The request did not specify what the “national security risk” would be, but the company believes that the reason for the government’s action is the possibility of a Fable 5 jailbreak.

The practice consists of using prompt engineering or simulations to make an AI bot return something different from the way it was programmed. The technique is used to ignore safety rules and ethical principles, perform restricted actions or implement a different bias in the models than initially programmed.

In a statement, Claude’s owner said that, despite complying with the order, she believes that the government’s directive is a “misunderstanding”. Anthropic disagrees that the possibility of a jailbreak is reason enough to prevent millions of people from using the tool.

“If this mandate were applied industry-wide, virtually all new implementation, at least from major model providers, would come to a halt,” the company wrote in a statement.

She herself had already informed about the possibility of \”invasion\” to the American government, but criticizes the attitude.

“As we have already stated, we believe that the government has the autonomy to stop risky implementations as part of a legal process, with transparency, fair, clear and based on technical facts. This action does not comply with these principles”, added the company.

The order came just three days after the models were released; Fable 5 and Mythos 5 were made available to the public last Tuesday, the 9th.

They would be available free of charge for just 13 days. After Monday, June 22, users used the model not under subscription plans, but via API pricing, where they would be charged for each request.

But this is not the first time that the American government and Anthropic have disagreed. In late February, the government ordered American agencies to stop using the startup’s AI models after the company refused, for ethical reasons, to allow the country’s armed forces to use the company’s models without restrictions and without technical safeguards.