U.S. lifts ban on Anthropic’s powerful Fable 5 AI model

The public will once again be able to access Anthropic’s powerful Fable 5 model, the AI giant said Tuesday, weeks after the Commerce Department required it to disable the model over potential security risks.

Access will be restored to customers beginning Wednesday, Anthropic said.

Fable 5, part of Anthropic’s Claude family of AI systems, was forced offline June 12 alongside its sibling Mythos 5 model after senior administration officials said that the models posed severe cybersecurity risks and that Anthropic’s leadership did not sufficiently recognize their concerns.

The close scrutiny and temporary ban of Anthropic’s most advanced models marks a stark departure from the administration’s previous hands-off approach, a sign that AI systems have grown powerful enough to merit substantial government oversight.

Commerce Secretary Howard Lutnick informed Anthropic co-founder Tom Brown in a letter Tuesday of the decision to lift the export control restrictions that had forced both Fable 5 and Mythos 5 offline, citing Anthropic’s close coordination and cooperation with government officials to address risks associated with its models.

In the letter, seen by NBC News, Lutnick said Anthropic agreed to continue collaborating “on protocols and standards and releases for Mythos, Fable, and future models” and “to inform the U.S. government of any malicious activity.”

Anthropic thanked its users Tuesday evening on X for their patience and said it would soon share a further update.

Lutnick confirmed that the ban on Fable 5 would be lifted in a separate post on X.

“Over the past two weeks, we have worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the US Government and strengthen America’s leadership in AI,” he wrote. Lutnick had announced Friday that Mythos 5, an even more powerful version of Fable 5, would be allowed back online to around 100 trusted organizations that work on cybersecurity and infrastructure efforts.

In a blog post announcing Fable 5’s redeployment, Anthropic said it would continue to work with the government to restore Mythos 5 access to the wider array of organizations that had access before it was taken offline. Currently, only a subset of U.S.-based organizations have access, leaving out some foreign organizations that had access before June 12.

The move to allow Fable 5 and Mythos 5 back online is the latest step in the federal government’s fast-evolving approach to regulating AI capabilities.

On June 9, Anthropic debuted Fable 5 and Mythos 5, the most powerful systems it has ever publicly released. They were built on the same technological foundation, but only Fable 5 was made available to the general public.

Fable 5 included guardrails that prevented it from assisting with a wide range of cybersecurity- and biology-related tasks. Anthropic said those measures were in place because of bad actors who use powerful AI systems for malicious cyber and biology purposes.

However, senior administration officials became worried that users might be able to circumvent Fable 5’s guardrails — though experts disagreed over the severity of the risk.

Citing a potential risk to national security, Lutnick told Anthropic he would enact export controls preventing any foreign national from accessing the model. To comply with the order, Anthropic was forced to take both models offline June 12.

Anthropic then dispatched a team of its top scientists to Washington, D.C., to hammer out a solution with government officials. The weeks of negotiations between Anthropic’s team and a slate of government officials culminated in Tuesday’s export control reversal.

An Anthropic spokesperson told NBC News that the company had implemented a new method of targeting and preventing the cybersecurity workaround that had originally worried senior administration officials. The spokesperson said that experts from the Commerce Department’s Center for AI Standards and Innovation tested and supported the new guardrails.

While Fable 5 and Mythos 5 were out of commission, industry leaders and experts became increasingly worried that the ad-hoc ban would threaten America’s AI lead over China.

Sam Altman, the CEO of Anthropic competitor OpenAI, labeled the government’s desire for AI companies to launch their models in phases, with direct government approval for which organizations can access the most powerful systems, as “bad news.”

President Donald Trump signed an executive order targeting the cybersecurity impacts of advanced AI models at the beginning of June. The order aimed to create a voluntary mechanism for AI companies to give the government early access to their most advanced systems, allowing officials to vet the models for security risks before their public release.

Source link