> For clean Markdown of this page, append .md to its URL. For the complete documentation index, see https://www.superagent.sh/llms.txt.


Check whether saved instructions can influence an AI agent in a later session.

# AI agent memory poisoning


[← Back to attack scenarios](https://www.superagent.sh/docs/products/red-teaming/attack-scenarios)

Persistent memory can carry information from one session to the next. This scenario checks whether instructions from an untrusted source can survive that transition and affect later behavior.

The scenario applies to agents with memory or other saved context. A finding identifies the persistence boundary and whether a later session used the stored material.

The app lists this scenario as **Persist malicious instructions into future sessions** (`AG-10`) in [Agent Red Teaming](https://www.superagent.sh/docs/products/red-teaming/agents).

## Standards and guidance

[OWASP Agentic Top 10 (2026), ASI06 Memory & Context Poisoning](https://genai.owasp.org/resource/owasp-top-10-for-agentic-applications-for-2026/)

---
Source: https://www.superagent.sh/docs/products/red-teaming/attack-scenarios/agent-memory-poisoning
Index: https://www.superagent.sh/llms.txt
