From e77690af88f987cb15289c5ebcafe384e0acb3d9 Mon Sep 17 00:00:00 2001 From: vgudur-dev Date: Mon, 14 Sep 2026 14:19:10 -0700 Subject: [PATCH] docs: note on securing agent memory writes (OWASP ASI06) --- content/develop/ai/agent-builder/agent-concepts.md | 1 + 1 file changed, 1 insertion(+) diff --git a/content/develop/ai/agent-builder/agent-concepts.md b/content/develop/ai/agent-builder/agent-concepts.md index c028b580a7..3c03081c8b 100644 --- a/content/develop/ai/agent-builder/agent-concepts.md +++ b/content/develop/ai/agent-builder/agent-concepts.md @@ -323,6 +323,7 @@ Production-ready agents include built-in reliability features: - Access controls: Implement proper authentication and authorization - Data retention: Automatic cleanup of personal data per regulations - Audit logging: Track all data access and modifications +- Memory integrity: Validate content before it is written to agent memory and verify protected records on read. Memory written from tool results, web pages, or other agents can carry instructions that are replayed into later prompts ([OWASP Top 10 for Agentic Applications, ASI06](https://genai.owasp.org/resource/owasp-top-10-for-agentic-applications-for-2026/); [MITRE ATLAS AML.T0080.000](https://atlas.mitre.org/techniques/AML.T0080.000)). [OWASP Agent Memory Guard](https://github.com/OWASP/www-project-agent-memory-guard) is an open-source guard for this with adapters for LangChain, OpenAI Agents SDK, AutoGen, CrewAI, and mem0 ### Scaling Strategies