THE CRUNCH
Meta has apparently fixed a quirk in its Muse AI model that previously hid its internal file system from users. The system now willingly reveals its contents when asked, a feature Meta says was the intended design. This follows a period where the filesystem was only accessible through accidental prompts, and Meta clarified that the exposure was not a bug.
The Verge reports that Meta has adjusted Muse to expose its internal file system by default. Previously, users had to discover the files through accidental prompts, and the model itself would warn that the information was not meant to be seen. Meta clarified that this new, more open behaviour is the intended design.
Meta's David Singleton explained on X that the decision to expose the filesystem was deliberate. He stated that this was a very deliberate choice, framing the feature as a core part of the model's design rather than a fix for a security issue.
This change comes shortly after Amazon blocked Meta's Muse AI agent from accessing its online store. Amazon cited privacy and security concerns, noting that the agent failed to identify itself and appeared to capture customer credentials. Meta did not notify Amazon that Muse would access its store.


