Anthropic Suspends AI Model Following US Security Concerns
Anthropic has suspended its advanced new AI model shortly after its public launch due to security concerns raised by US authorities.
In an official statement on its website, Anthropic announced it was instructed to restrict access to Claude Fable 5, an AI program the company described as "too powerful," specifically preventing foreign nationals from using it.
"The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance," the company wrote.
Anthropic is also engaged in an ongoing legal dispute with the Trump administration regarding an order prohibiting government agencies from using its AI tools. The BBC has sought comment from the US Department of Commerce.
Claude Fable 5 is a variant of Anthropic's Claude Mythos, an AI system competing with OpenAI's ChatGPT and Google's Gemini.
Anthropic stated that US national security officials have not specified particular concerns.
"Our understanding is that the government believes it has become aware of a method of bypassing, or 'jailbreaking' Fable 5," the company said.
Jailbreaking refers to circumventing software restrictions designed to protect cyber networks, potentially allowing unauthorized access to sensitive data or enabling restricted features.
Anthropic reviewed a demonstration of this jailbreaking technique, which identified a small number of previously known, minor vulnerabilities.
"These vulnerabilities all appear relatively simple, and we have found that other publicly-available models are able to discover them as well without requiring a bypass."
Before releasing Claude Fable 5, Anthropic highlighted various safeguards implemented to prevent cyber attacks.
Leaders in finance, technology, and government had expressed concerns about the public rollout, which followed a private release in April used to preview and test vulnerabilities within the system.
Anthropic granted pre-release access to a limited number of organizations due to the tool's intelligence and potential risks related to exploiting or hacking computer systems.
The company described Claude Fable 5 as "too powerful to release" prior to its public launch, a claim some critics viewed as exaggerated marketing.
"Fable's capabilities exceed those of any model we've ever made generally available," the company said.
Expert Perspectives on Access Restrictions
Gina Neff, Professor of Responsible AI at Queen Mary University London, told the BBC that limiting access to the model could hinder the development and safe testing of AI systems.
"We're in uncharted territory at this point," she said.
"People within the AI industry have been warning us that these tools are getting better very rapidly and that we have to be able to build up capabilities to keep our companies safe from cyber attacks."
Neff noted that the UK government's AI Security Institute found in tests that the model could exploit defenses and systems 73% of the time.
"It's a step change in capability in cyber security," she added.
Anthropic's Legal Challenges and Government Relations
Anthropic has recently faced scrutiny from the Trump administration.
Former President Donald Trump publicly criticized the company, and then US Defence Secretary Pete Hegseth labeled it a "supply chain risk," marking the first time a US company received such a designation.
This designation indicates that a tool or service is deemed insufficiently secure for government use and is typically reserved for companies based in adversarial countries.
Anthropic is suing the Pentagon over this designation. A US judge ruled that the Pentagon's directive could not be enforced, allowing government agencies and organizations working with the US military to continue using Anthropic's services while the lawsuit proceeds.
