Anthropic Halts Global Access to Advanced AI Models Following US Export Control Order

New Delhi: Artificial intelligence company Anthropic has suspended access to its advanced AI models, Fable 5 and Mythos 5, for all foreign nationals after receiving an export control directive from the United States government.

In a statement posted on its blog, the company said the order requires it to block access to the two models for any foreign national, including Anthropic employees, regardless of whether they are located inside or outside the United States.

To comply with the directive, Anthropic said it is immediately disabling the models for all affected customers.

The company, however, expressed disagreement with the government's decision, arguing that the action appears to be based on the discovery of a limited jailbreak technique capable of bypassing some model safeguards.

According to Anthropic, applying such a standard across the AI industry could significantly disrupt the deployment of new frontier AI systems.

The US government has not publicly detailed the specific national security concerns behind the directive. Anthropic said authorities informed the company about a method that could potentially circumvent certain protections built into Fable 5.

After reviewing the demonstration, the company concluded that the technique exposed only a small number of previously known and relatively minor vulnerabilities. It also noted that similar weaknesses could reportedly be identified using other publicly available AI models without requiring any special bypass methods.

Anthropic highlighted that before launching Fable 5, it conducted extensive safety testing in collaboration with US government agencies, the UK's AI Safety Institute (AISI), independent security organisations and its own internal teams.

The company claimed that thousands of hours of red-team testing showed Fable 5's safeguards to be more robust than those of earlier-generation models.

Anthropic further stated that no testing group has yet succeeded in discovering a "universal jailbreak"—a technique capable of broadly overriding the model's safety protections and enabling a wide range of restricted cyber capabilities.

The development comes amid growing scrutiny of advanced AI systems by governments worldwide, as regulators seek to balance technological innovation with concerns related to national security, cybersecurity and misuse of powerful AI tools.
 

With inputs from IANS

Advertisement