The US government has lifted an export ban on Anthropic's most advanced artificial intelligence tools, just weeks after ordering the company to restrict access to them over national security concerns. The company said it will begin restoring access to Claude Fable 5 and Mythos 5 on Wednesday after being notified that the Department of Commerce has removed restrictions on the two models.
The ban was abruptly imposed on 12 June over fears that the tools could be used by hackers to exploit weaknesses in computer systems. Both models had only been released three days earlier, on 9 June. Fable 5 is designed for the consumer market and is capable of deep reasoning and performing complex tasks independently. Mythos 5, aimed at businesses and cybersecurity experts, is said to be able to identify vulnerabilities in computer code and exploit them.
“US lifts export ban on Anthropic's advanced AI tools weeks after imposing it over hacking fears.”
In a letter seen by the BBC, Commerce Secretary Howard Lutnick wrote that Anthropic has addressed the risks. “Anthropic has agreed to proactively detect and address security risks associated with the models,” Lutnick said. The firm has also agreed to collaborate on future releases and alert the government of any malicious activity, he added. However, Lutnick noted that the department reserves the right to reconsider its decision if necessary.
Anthropic previously pushed back against the ban, arguing that the government had not pinpointed specific concerns. “Our understanding is that the government believes it has become aware of a method of bypassing, or ‘jailbreaking’ Fable 5,” the company said at the time, referring to slipping past software safety restrictions. “However, we disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people.”
The BBC has contacted the Commerce Department for further information. The lifting of the ban marks a dramatic reversal for Anthropic, whose Claude platform competes with OpenAI's ChatGPT and Google's Gemini.