docs(observability): add threat detection page

[eeee2345]

Adds a new docs page under spring-ai-docs/src/main/antora/modules/ROOT/pages/observability/threat-detection.adoc describing the runtime threat surface for Spring AI applications and pointing at open-source detection-rule infrastructure that downstream builders can plug into their deployments.

Why now

Three CVE family entries against Spring AI were disclosed on 2026-05-08:

• CVE-2026-41705 — MilvusVectorStore filter-expression injection
• CVE-2026-41712 — PromptChatMemoryAdvisor cross-conversation memory leakage
• CVE-2026-41713 — PromptChatMemoryAdvisor memory poisoning

All three are patched in Spring AI 1.0.0 and above. The added page documents the threat shape, sketches a minimal CallAroundAdvisor integration pattern for detection-rule loading, and lists multiple open-source detection-rule corpora as plural starting points (ATR, garak, HarmBench, OWASP A-S-R-H). No single-vendor endorsement.

Scope

• Adds one new file: pages/observability/threat-detection.adoc (80 lines)
• Adds one nav entry under existing observability/index.adoc in nav.adoc
• No source code changes
• No dependencies added

Honest about limits

The page explicitly documents what detection rules do not catch (paraphrase bypasses, multilingual coverage gap, configuration / authentication issues) and frames the layer as defense in depth rather than replacement for framework patches.

Maintainer voice

I am the maintainer of one of the referenced corpora (Agent Threat Rules, github.com/Agent-Threat-Rule/agent-threat-rules) — disclosed as a conflict of interest in the page text. The reference list is intentionally plural and includes peer projects to avoid single-vendor framing.

Happy to revise scope, structure, or tone per maintainer feedback. If a threat-detection section fits better as additions to existing pages (advisors.adoc / chat-memory.adoc / vectordbs.adoc) rather than a new page, I can rework.

Sources cited in the page

• Spring Security Advisories for the May 2026 CVE family
• Five Eyes "Careful Adoption of Agentic AI Services" joint guidance (2026-04-30)
• Microsoft MSRC "When prompts become shells" (2026-05-07)

[eeee2345]

Flagging this one — it is still labelled status: waiting-for-triage, so I do not think it has reached a reviewer yet rather than having been looked at and set aside.

It adds an observability docs page on threat detection. Docs only, mergeable against main.

Whoever handles observability docs triage: happy to adjust scope, move the page, or split it if the placement is wrong. And if agent threat detection is out of scope for the Spring AI docs, a close is a fine outcome — I would just rather have a direction than leave it in the triage queue indefinitely.

[sdeleuze]

Thanks for the ping, I prefer to decline such contribution to our documentation that seems a bit out of topics and looks too much AI generated (unless I am mistaken).