AI runtime security guides
Twelve guides explain AI runtime security for practitioners: what it is, how AIDR differs from EDR, inline versus out-of-band enforcement, failure modes, latency budgets, logging, RAG pipelines, metrics, cloud provider controls and open-source projects. Each claim carries a source and a 16 Nov 2026 read date.
Facts read on .
12 guides
- AI Runtime Security Explained: What AI runtime security is, which actions it stops, where it enforces and how it differs from posture management, with sources read on 16 Nov 2026.
- AI Runtime Security vs Guardrails vs Red Teaming: How AI runtime security, guardrails and red teaming differ: when each acts, what each produces and how they fit together, with sources.
- AIDR vs EDR vs XDR: What Each Watches: What AI detection and response (AIDR), EDR and XDR each watch, and how to run them together, with vendor documentation read on 16 Nov 2026.
- Cloud Provider Built-In Runtime Controls: What Amazon Bedrock Guardrails, Azure AI Content Safety and Google Model Armor include, and when a separate AI runtime security tool is needed.
- Fail Open or Fail Closed: What Vendors Publish: What happens when an AI runtime control fails, which vendors document a failure mode, and what to ask for, using documents read on 16 Nov 2026.
- Inline or Out-of-Band Runtime Enforcement: Inline enforcement can stop an AI action before it completes; out-of-band monitoring sees it afterward. How to choose, with vendor and cloud sources.
- Open-Source AI Runtime Security Projects: Open-source projects that enforce at runtime: NeMo Guardrails, Llama Guard, Prompt Guard, LLM Guard and Guardrails AI, compared with commercial tools.
- Runtime Protection for Customer-Facing LLM Apps: Attacks on a public LLM app and the controls that sit in the request path: input checks, output checks, data filters and logs, with sources.
- Runtime Security for RAG Pipelines: What goes wrong in a retrieval-augmented generation pipeline at runtime and where to place controls: ingest, retrieval, prompt, output.
- Runtime Security Latency Budgets for LLM Apps: How much delay an AI runtime control may add to an LLM app, how vendors state it, and how to measure it yourself, with sources read on 16 Nov 2026.
- Runtime Security Metrics: False Positives and Blocks: Which metrics show whether an AI runtime control works: blocked rate, false positives, misses and latency, and how to set a false positive budget.
- What to Log From Runtime Security Decisions: Which AI runtime security decisions to record, what fields to keep, and how to send them to a SIEM, with vendor documentation read on 16 Nov 2026.