Amazon security research reportedly led to the White House’s Anthropic Fable ban

Article excerpt

Amazon's internal security research appears to have triggered a White House export control directive that forced AI startup Anthropic to cut off access to its Fable 5 and Mythos 5 models. According to the Wall Street Journal, Amazon researchers demonstrated that carefully crafted prompts could extract information useful for cyberattacks from Fable 5, findings that CEO Andy Jassy discussed with White House officials. The result: Anthropic restricted access to the models, with Amazon declining to comment on its role in the episode.