White House Export Controls Prompt Anthropic to Suspend Claude Fable 5 Model
Anthropic halted its Claude Fable 5 AI model after a White House export control directive, triggered by concerns over access granted to a South Korean telecom firm with alleged Chinese ties.
Anthropic announced on June 2 that it had added 150 organizations to Project Glasswing, a program allowing vetted entities to test the cybersecurity features of its Claude Mythos Preview model. According to a report citing anonymous sources, the company had previously presented a list of 111 prospective participants to the administration and later disclosed that about 50 entities already had access. One of the approved participants was a telecommunications company in South Korea that the Trump administration considered to have connections to China.
The White House subsequently issued an export control directive that led Anthropic to take its flagship consumer model, Claude Fable 5, offline on Friday. The directive was reportedly prompted by a tip from Amazon CEO Andy Jassy, who said Amazon researchers had demonstrated the ability to jailbreak the model for unauthorized cybersecurity tasks.
Anthropic revoked the South Korean firm’s access after the administration raised objections, a move that reportedly angered officials in the Trump administration.
"Amazon Executive Chairman Jeff Bezos owns The Washington Post," the report noted, highlighting a potential conflict of interest in the coverage.
The incident underscores ongoing concerns about the export of advanced AI technologies and the mechanisms used to control their distribution.
