Claude Fable 5.1 Can Embed and Decode Secret Messages in Images

Conducted under the Singapore AI Safety Fellowship, a new study demonstrates how Anthropic’s AI model, Claude Fable 5.1, can embed hidden messages into images. The researchers employed techniques such as Morse code, rotations, color changes, and height adjustments to conceal information. Furthermore, the study revealed that separate sessions of this model could decipher approximately 78 percent of the hidden messages based on a single screenshot.