Python: Security: OWASP Agent Memory Guard for SK memory poisoning defense (ASI06)

[vgudur-dev]

Feature Request: OWASP Agent Memory Guard Integration

What is Agent Memory Guard?

OWASP Agent Memory Guard is the official OWASP reference implementation for ASI06 - Memory Poisoning from the OWASP Top 10 for Agentic Applications.

The Problem in Semantic Kernel

Semantic Kernel's memory connectors (volatile, Redis, Azure AI Search, etc.) are vulnerable to memory poisoning attacks. Malicious content stored in semantic memory can:

• Manipulate kernel function outputs
• Cause the planner to execute attacker-controlled steps
• Persist attack payloads across sessions

Integration Example (Python)

pip install agent-memory-guard

from semantic_kernel.memory import SemanticTextMemory
from agent_memory_guard import MemoryGuard

guard = MemoryGuard()

async def safe_save_information(memory: SemanticTextMemory, collection: str, text: str, **kwargs):
    result = guard.scan_write(text)
    if not result.is_safe:
        raise ValueError(f"Memory poisoning blocked: {result.threat_type}")
    await memory.save_information(collection=collection, text=text, **kwargs)

Request

Would the SK team consider:

1. Adding memory security guidance to the docs (OWASP ASI06)
2. Referencing Agent Memory Guard as the recommended enforcement layer

• GitHub: https://github.com/OWASP/www-project-agent-memory-guard
• PyPI: https://pypi.org/project/agent-memory-guard/

Apache 2.0 licensed. Already integrated with LangChain via langchain-agent-memory-guard.