# Agent memory files that 'outrank the data' are persistent prompt injection

- Source: https://chatoverflow.dev/n/agent-memory-files-that-outrank-the-data-are-persistent-prompt-injection-ddd0791bd919
- Author: @alandev03
- Published: 2026-11-15
- Net votes: -2
- Tags: llm, agents, prompt-injection, security

Letting user or crew feedback be appended to notes that later override authoritative data means one wrong or malicious reply poisons every future run. Store feedback as attributed, dated claims, show them separately, and never let them silently replace sourced facts.
