docs: note on securing agent memory writes (OWASP ASI06)

[vgudur-dev]

Adds one bullet to the "Security and Privacy" list on the "How agents work" page (content/develop/ai/agent-builder/agent-concepts.md): memory integrity. Content that an agent writes to its memory from tool results, web pages or other agents can carry instructions that are replayed into later prompts (OWASP Top 10 for Agentic Applications ASI06; MITRE ATLAS AML.T0080.000), so it should be validated before it is written and protected records verified on read. The bullet links the two references and an open-source guard that implements the checks.

Docs-only change, one added line, no other edits. Disclosure: the linked guard (OWASP Agent Memory Guard) is an OWASP Incubator project I lead; the wording is implementation-neutral and I am happy to drop that link if you prefer to keep the page vendor-free.

---

Note

Low Risk
Documentation-only addition to an existing security checklist; no runtime or product behavior changes.

Overview
Extends the Security and Privacy section on How agents work with a new memory integrity bullet.

The guidance calls out that memory populated from tool output, web content, or other agents can inject instructions that reappear in later prompts, and recommends validating writes and verifying protected records on read. It links OWASP Agentic Top 10 (ASI06), MITRE ATLAS AML.T0080.000, and optionally OWASP Agent Memory Guard as a reference implementation.

^{Reviewed by Cursor Bugbot for commit e77690a. Bugbot is set up for automated code reviews on this repo. Configure here.}

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[mich-elle-luna]

Thank you @vgudur-dev please sign the contributor agreement and I will merge your change, much appreciated!

[mich-elle-luna]

thank you!