Runtime security topic hub
AI agent runtime security articles for live action checks and audit evidence
Use this hub for the live-control layer: what to check before a tool executes, what to inspect after a tool returns, how policy becomes runtime enforcement, and what audit evidence should remain.
· 9 articles · 3 guides
Search questions this hub answers
What is AI agent runtime security?
What are pre-tool call and post-tool response checks?
How do runtime controls prevent agent data exfiltration?
What audit evidence should an AI agent security layer keep?
AI Security Posture Management (AI-SPM): What It Covers and Where Runtime Takes Over
AI security posture management explained: agent and tool inventory, credential mapping, configuration baselines, posture drift detection, and the runtime gap that posture scans cannot see.
Guardrails, Sandboxes, and Gateways: AI Runtime Security Compared
Three control families claim the "AI runtime security" label. What inline guardrails, OS sandboxes, and gateways each catch, what each is structurally blind to, and how to combine them.
Prompt Injection Detection: How Pattern Matching, Classifiers, and LLM Judges Compare
Prompt injection detection methods compared: deterministic pattern checks, ML classifiers, and LLM judges - the recall, false-positive, and cost trade each makes, and how to evaluate a detector honestly.
Claude Code Security: A Practical Hardening Guide for Local Coding Agents
Six controls that harden Claude Code and other local coding agents: untrusted-input discipline, least privilege, config-write protection, credential scoping, egress limits, and audit evidence.
AI Agent Security Monitoring: Audit Logging, Runtime Telemetry, and Forensics
What an AI agent audit log should actually capture — prompts, tool calls, results, policy decisions — how to keep it tamper-evident, and how to reconstruct an incident from the evidence after something goes wrong.
AI Agent Data Exfiltration: Prompt Injection, Tools, and Runtime Defense
How agents leak data: an attacker plants instructions in content the agent reads, the agent retrieves something sensitive, then sends it out through a legitimate tool. What each defense layer can and cannot see along that path.
Agentic Ransomware: What JADEPUFFER Means for AI Agent Security
JADEPUFFER and clean-repo coding-agent exploits show why AI agent security needs runtime action checks, credential boundaries, egress controls, and audit evidence.
AI Agent Runtime Security Checklist
A practical checklist for securing AI agents at runtime: inventory tools, scope credentials, inspect inputs, check tool dispatch, screen tool results, verify outputs, keep audit evidence, and roll back safely.
MCP Tool Poisoning: Runtime Defense for AI Agents
MCP tool poisoning explained: poisoned tool descriptions, schemas, metadata, and results; where scanners help; and why runtime dispatch and result checks still matter.
Guides for this topic
Guide
How to Detect Prompt Injection Attacks in AI Agents
Practical guide to direct and indirect prompt injection, tool-using agents, runtime inspection, and where Clawmont fits.
Guide
MCP Security Best Practices for AI Agent Deployments
MCP server trust, tool authorization, input validation, runtime monitoring, credential scoping, and audit logging.
Guide
AI Agent Security Tools Compared (2026)
Neutral comparison of model safety, classifiers, runtime inspection, API gateways, sandboxing, and supply-chain scanning.
Primary source page
For a concise definition and product-level framing, start with AI agent runtime security definition. The articles above expand the surrounding search intents and link back to the canonical Clawmont topic page.