“Andrew Morris of GrayNoise Intelligence stated that Amazon's researchers did not present evidence that they had successfully prompted Anthropic's Fable model to produce functional exploit code.”