dify - 💡(How to fix) Fix Security: OWASP ASI06 memory poisoning defense for Dify agent memory

Memory Poisoning Defense for Dify Agents (OWASP ASI06)

Hi Dify team 👋

Dify is one of the most widely deployed agent application platforms. As agents built on Dify handle increasingly sensitive workflows, ASI06 — Memory Poisoning (from the OWASP Top 10 for Agentic Applications 2025) is an emerging threat that deserves attention.

The Threat

Malicious content injected into an agent's memory (conversation history, knowledge base, tool outputs) can persist across sessions and silently manipulate future agent behavior — bypassing all LLM-level safety filters.

Proposed Integration

OWASP Agent Memory Guard is an official OWASP reference implementation for detecting and blocking memory poisoning attempts.

from agent_memory_guard import MemoryGuard

guard = MemoryGuard()
result = guard.scan(memory_content)
if result.is_safe:
    memory_store.write(memory_content)

Install: pip install agent-memory-guard

Ask

Would the Dify team be open to:

1. A security note in the memory/knowledge base configuration docs?
2. An optional memory validation hook in the agent pipeline?

Happy to draft a docs PR. This would make Dify one of the first enterprise agent platforms with a documented ASI06 mitigation.