Amazon’s internal security research reportedly prompted the White House to impose export controls on Anthropic’s Fable 5 and Mythos 5 models, according to the Wall Street Journal. The directive followed conversations between Amazon CEO Andy Jassy and government officials after the e-commerce giant demonstrated the models' potential for malicious use.
The research paper indicated that through a series of carefully crafted prompts, Amazon’s team could coax Fable 5 into revealing information useful for cyberattacks. This finding escalated concerns about AI safety and national security, leading to swift government action. The block now prohibits foreign nationals from accessing the models.
Anthropic subsequently cut off access to the models, complying with the directive. The Verge reports the situation is complicated by the fact that many foreign researchers and developers rely on these tools. Amazon has not yet responded to requests for comment on the matter.
The episode underscores the growing tension between open AI development and security imperatives. Critics argue that such broad restrictions could stifle innovation and harm global research collaborations, while supporters insist they are necessary to prevent malicious actors from weaponizing advanced AI.
Some experts caution that the evidence linking prompt-based exploits to real-world threats remains thin. They warn that overreaction could set a precedent for censorship, making it harder to audit and improve safety in frontier models.