Anthropic says US has lifted export controls on Fable and Mythos AI models after security fears | AI (artificial intelligence)

Anthropic has restored customer access to its latest powerful AI model, Fable, after more than two weeks of outage stemming from security concerns from the US government that it could be exploited to enable serious cyberattacks.

The San Francisco company said export controls had been lifted, citing a social media post by US commerce secretary Howard Lutnick: “Over the past two weeks, we have worked closely with Anthropic to analyze and certify Fable 5 to ensure compliance across the US Government and strengthen America’s leadership in artificial intelligence.”

Three days after the release of Claude Fable 5 on June 9, the US government ordered Anthropic to immediately restrict access to this model and the related model, Claude Mythos 5, and prohibit their use by foreign nationals within or outside the US. Washington feared that powerful artificial intelligence could be used by foreign adversaries to launch devastating cyberattacks.

On Friday, the US government partially reversed that order by allowing Anthropic to release Mythos 5 to certain “trusted” US outlets. Mythos has only been released to a small number of trusted partners for use in defensive cybersecurity.

“Anthropic has agreed to proactively detect and address security risks associated with the models; to work diligently with the US government on protocols, standards and releases for Mythos, Fable and future models; and to notify the US government of any malicious activity,” Lutnick said in a letter to Anthropic seen by Reuters.

The ban was seen as a hardening of the White House’s stance on regulation of AI companies in the US; This stance has previously emphasized a light touch, underpinned by a desire to beat China in the so-called race for AI dominance.

The US government took action after being informed of a report that Amazon researchers had found a method to find Fable 5’s vulnerabilities, showing how such a weakness could potentially be exploited by a malicious cyber attacker.

Anthropic responded to the export control by pulling the plug entirely, but said it was now increasing safeguards. A blog post stated: “Researchers from the U.S. Department of Commerce Artificial Intelligence Standards and Innovation Center (CAISI) tested both our previous and new protections and found them to be exceptionally strong.”

In a statement late Tuesday, Anthropic said it would “increase our government collaboration” and “create dedicated Anthropic teams to work on shared government priorities.”

Last week, intelligence agencies in Australia, the US, the UK, New Zealand and Canada warned that while AI “will help us improve cyber defenses over time, it is also accelerating the speed, scale and sophistication of cyber threats.”

The warning made by Five Eyes agencies stated: “Frontier AI models are expected to fundamentally transform both offensive and defensive cyber capabilities, exceeding current industry expectations. The timeline is months, not years.”

Susie Wiles, Donald Trump’s chief of staff, shared: “I express my gratitude to companies across industries who continue to work closely with the White House to implement the President’s EO: ‘Promoting Advanced AI Innovations and Security.’ This includes excellent work on advanced model access, guardrail testing and safety.

“Government and the private sector have worked together in a way we’ve never seen before, and this foundation of America First is unprecedented. Our shared priority remains: ensuring the best technology is deployed as quickly and safely as possible.”

But the US government’s review of which companies have access to the models has sparked criticism.

OpenAI chief executive Sam Altman said last week that extensive security testing “is not a bad idea. I just don’t like the idea of ​​the government picking customers.” The ChatGPT creator has delayed the full public launch of GPT-5.6 at the request of the US government, limiting its access to a small group of vetted partners.