microsoft / microsoft/semantic-kernel
Python: Security: OWASP Agent Memory Guard for Semantic Kernel memory protection
Nobody has claimed this yet.
- Dominant language
- C#
- Stars
- 28.6k
- Forks
- 4.8k
- Avg merge
- 14h 13m
- Merged PRs (30d)
- 18
Description
Feature Request: Memory Security Integration
Hi Semantic Kernel team,
Semantic Kernel is Microsoft's premier framework for building AI agents and copilots. As SK agents increasingly use memory stores (volatile, semantic, episodic), memory poisoning is an emerging security concern for enterprise deployments.
The threat: Malicious content injected into SK memory stores can silently bias agent behavior across sessions — a persistent, hard-to-detect attack vector.
OWASP Agent Memory Guard is the reference implementation for OWASP ASI06 (Memory Poisoning) in the OWASP Top 10 for Agentic Applications:
- Runtime scanning of memory reads/writes for injection payloads, secret leakage, and structural tampering
- SHA-256 integrity baselines to detect post-write tampering
- Policy-driven enforcement (block / warn / strip)
- Sub-100μs overhead
pip install agent-memory-guard
GitHub: https://github.com/OWASP/www-project-agent-memory-guard
PyPI: https://pypi.org/project/agent-memory-guard/
As a fellow Microsoft engineer working on AI security, I'd love to explore adding this as a recommended security layer in SK's memory pipeline or as a featured integration in the SK documentation.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by reviewing Semantic Kernel's memory pipeline and the linked OWASP Agent Memory Guard project and PyPI package. Determine whether the intended outcome is a recommended integration or a documentation entry, then define acceptance criteria for memory scanning, integrity checks, and block, warn, or strip policies.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- ai, security
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Quiet
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100