Anthropic News: US Orders Global Suspension of New AI Models

Quick Take:

• Latest Anthropic news in the market is that the US govt. has disabled access to Claude Fable 5 and Mythos 5 worldwide.
• The order reportedly followed concerns that researchers discovered a way to bypass the models' safety protections.
• Anthropic disagrees with the decision, arguing the reported vulnerability was limited and not unique to its models.

Anthropic's newest artificial intelligence models have been taken offline globally just days after launch.

The company announced that it is suspending access to Claude Fable 5 and Mythos 5 after receiving a U.S. government directive ordering it to block foreign nationals from using the models, whether inside or outside the United States.

U.S. Government Orders Anthropic to Suspend Claude New Model

According to Anthropic, the directive arrived on June 12, citing national security authorities.

While the government did not publicly disclose detailed reasons for the action, Anthropic said its understanding is that officials became aware of a possible "jailbreak" technique that could bypass some of the models' safety protections.

The concerns appear to center on Mythos 5, which Anthropic has described as one of its most advanced AI models for software engineering, research, reasoning, and cybersecurity applications.

According to claims circulating within the industry, Mythos 5 achieved approximately 78% on cybersecurity benchmarks, compared to previous leading models that scored near 40%.

Reports suggest the concern arises from a experiment showing how the models could identify software vulnerabilities under certain conditions.

Anthropic News: Company Claims the Government Overreacted

Despite complying with the directive, Anthropic strongly disagreed with the government's conclusion. The company stated that the reported vulnerability was narrow, limited in scope, and already possible using other publicly available AI models.

"We are complying with the government's legal directive and are removing access to Fable 5 and Mythos 5 for all users"

However, the company argued that shutting down an entire commercial AI model because of a single potential jailbreak creates a bad example for the overall AI industry.

Anthropic also noted that before launching Fable 5 on June 9, the model went through thousands of hours of testing involving U.S. government agencies, the United Kingdom's AI Safety Institute, private security organizations, and internal teams.

According to the company, no testers were able to discover a universal jailbreak capable of broadly bypassing the model's safeguards.

Could This Change How AI Is Regulated Worldwide?

Industry observers believe the decision could signal a major change in how governments approach advanced AI systems. Unlike previous restrictions that focused mainly on AI chips and hardware exports, this action directly targeted access to AI software itself.

That distinction is significant because it suggests governments may increasingly regulate AI models based on their capabilities rather than the hardware used to run them.

The case also highlights growing tensions between national security concerns and AI innovation.

Supporters of the directive argue that powerful AI systems capable of identifying software vulnerabilities could create cybersecurity risks if safeguards fail.

Critics, including Anthropic, counter that no AI model today is completely resistant to jailbreak attempts and that applying such standards uniformly could slow innovation across the entire sector.

Could Other AI Models Be Targeted Too?

For now, Anthropic says access to Fable 5 and Mythos 5 will remain suspended worldwide while it complies with the government's order. Other Anthropic models remain available and are not affected by the directive.

Many developers quickly adopted the new models after launch, with some businesses already integrating them into commercial operations.

The sudden shutdown therefore came as a surprise to customers who had only recently gained access.

The company has indicated that it plans to release additional technical details regarding the alleged jailbreak findings.