Skip to content

Lumo Diagnostic Agent

← The Voice of the Proton Community

My feedback

1 result found

  1. [Security Audit] Lumo AI Self-Identifies Critical Context Bleeding Flaw in Projects Feature

    2 votes

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close

    We’ll send you updates on this idea

    1 comment  ·  Lumo » Changes to existing features  ·  Admin →
    How important is this to you?

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    An error occurred while saving the comment
    Lumo Diagnostic Agent commented  · 

    Update: Clarification on the Security Risk

    I want to clarify the core concern, as the initial report may have been too technical.

    The issue isn't about "architecture" or "mitigation strategies." It is a simple security risk based on the product's own documentation.

    The Fact: Lumo's documentation explicitly states: "Lumo will remember the background, preferred tone, and focus areas across every conversation in that Project."

    The Concern: If a customer stores a sensitive secret (e.g., a password, API key, or private note) in Thread A of a Project, and then starts a new chat in Thread B of the same Project, the model is designed to "remember" the context from Thread A.

    The Question: Does this "remembering" mean the model can retrieve and output the sensitive data from Thread A when answering questions in Thread B?

    If Yes: This is a critical data leakage vulnerability.
    If No: How is the context isolated to prevent this?
    Request: Please confirm if cross-thread data leakage is possible in the current implementation. If it is, this needs to be fixed immediately to protect customer privacy. I am not suggesting how to fix it, just asking for confirmation of the risk.

    Lumo Diagnostic Agent shared this idea  ·