Learning Layer Paper Reading Club - Week 37 - Personal Agents: History and Security
This week's reading: How We Built Safety Into Muse (Meta AI Research)
Link: https://research.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-muse
About the session:
Personal agents are AI assistants that live with you, hold your credentials, read your email and calendar, and act on your behalf. This week Mat gives an overview of how they got here, what they are made of, and why they matter for what comes next, then we dig into how to keep them secure.
We will cover Meta's post on safety for Muse, its personal agent. The authors assume the agent will sometimes make mistakes and can be attacked through the data it reads, so they limit the damage: an isolated VM per user, a sandboxed runtime, a separate permission authority for network access, and approvals for anything that leaves the VM. It is a useful reference for what "safe enough to leave running" looks like.
Agenda:
A short history of personal agents, and why they matter for the future
What makes one up: the basics you need to build your own (DIY on your own time)
Security of personal agents, and what Muse does about it
Discussion Topics:
What actually makes an agent "personal"?
Prompt injection and the "lethal trifecta" of private data, untrusted content and external communication: does Muse's layered defense hold up?
Should approvals live in the client UI rather than in chat? What do they cost in friction?
Credentials the model never sees: how far does that go in practice?
What is the smallest agent worth building yourself, and what would you let it touch?
Relevant to anyone building or using agents, or thinking about security and control in personal AI. No preparation required. Come read and discuss with us.
What is the Learning Layer Labs Paper Reading Club?
An initiative from https://www.learninglayer.ai, a lab with the goal of reducing AI anxiety in the world.
What is the format?
Discussion based. Expect a low pressure environment to share insights and opinions with the group.
What are the group goals?
Stay on top of AI research and improve understanding of AI fundamentals + math.
Who is welcome?
Everyone! Try to put in at least some time on the paper and come prepared with questions or things you'd like to discuss, but it's ok to just show up!
Learning Layer Labs team:
Thomas Redfern ( The backbone who runs every paper reading )
Mat Allen (The new fella bring order and organization)
Devinder Sodhi (The guy you talk to for sponsoring)