Evidence · DiggingBeagle record
Role Grounding Attack: quoted dialogue contaminating agent preference memory
Original study of role laundering from quoted third-party dialogue into an agent’s apparent user-preference memory.
- Published
- Oct 30, 2026
- Source role
- primary disclosure
Evidence record
Original study of role laundering from quoted third-party dialogue into an agent’s apparent user-preference memory.
Claim-level citations (1)
- supportsQuoted third-party dialogue became a false user preference in agent memory tests: The Role Grounding Attack study documents conditions in which quoted external dialogue is elevated into an agent’s representation of a user preference.
arXiv:2610.38209, attack formulation and memory update evaluation