Anthropic has suspended access to its most advanced artificial intelligence (AI) models in response to a United States Government order based on national security concerns.
The privately-owned but soon-to-be-listed AI model developer said it received the export-control directive on Friday, requiring it to suspend access to its newly launched Fable 5 and Mythos 5 models for foreign nationals.
Anthropic said the broad scope of the order left it with little choice but to disable the models for all users to ensure compliance.
“The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers," the company said in a statement.
Access to its other AI models, including the well-known Claude remained unaffected.
The Government has not detailed the concerns, but Anthropic said it understood officials believed they had identified a way of bypassing Fable 5’s safeguards, known as a ‘jalbreak’, to obtain information related to cybersecurity vulnerabilities.
But it also said the vulnerabilities uncovered were minor, previously known and could also be identified by other publicly available AI models.
"We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people," Anthropic said.
The company’s relationship with the Government broke down this year after it refused to allow the U.S. military to use its AI models for domestic surveillance and fully autonomous weapons systems, resulting in it being put on a supply chain blacklist.
Just as that dispute looked like it was improving, the new order was put in place.
Anthropic said the action did not adhere to the principles that the statutory process under which the Government can block unsafe deployments was transparent, fair, clear and grounded in technical facts.
“We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers,” it said.
Earlier this month, the company moved closer to a stock market listing by filing for an initial public offering through the submission of a draft registration statement to the United States’ Securities and Exchange Commission (SEC).
It was valued at US$965 billion (A$1.4 trillion) in a capital raising earlier this year.
