@jonny as you said elsewhere in the thread ... i'm not sure about the business model here.
But it really does seem to be by design. There's an architecture diagram and a fair number of details in https://research.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-muse
Oh no, you got me to not only read a post on Meta's site but actually share a link to it. You really are an evil genius!