# ASI06 - Memory & Context Poisoning

> Attackers plant malicious data in agent memory or RAG databases to influence future decisions.

- **Category**: Owasp Agentic Ai
- **Subcategory**: top10_2026
- **Canonical URL**: https://designpattern.fyi/owasp_agentic_ai/asi06_memory_context_poisoning/

---

## Description
'**Intent**: Protect agent memory and context stores from adversarial manipulation that could influence future behavior.

**Context**: Agents with persistent memory or access to shared knowledge bases can be poisoned by injecting malicious data that influences future reasoning and decision-making.

**Solution**: Validate all data before memory storage. Implement memory integrity checks. Use access controls on shared memories. Monitor for anomalous memory modifications. Apply data provenance tracking. Implement memory hygiene policies.'



## Use Cases
Use when agents have persistent memory, shared knowledge bases, or retrieval-augmented generation capabilities.





## Trade-offs


### Advantages

- Maintains reasoning integrity

- Prevents long-term manipulation

- Enables memory auditing

- Supports reliable agent behavior




### Considerations & Drawbacks

- Memory validation adds overhead

- Subtle poisoning is hard to detect

- Memory pruning may lose useful data







---
**Reference**: [Original Source](https://genai.owasp.org/resource/owasp-top-10-for-agentic-applications-for-2026)