Blog
AI Agent Security
from the inside.
Technical deep dives on the attack surfaces that matter when AI agents act autonomously - and how to defend against them.
We started publishing in mid-2026 and keep this archive deliberately small - every post here has to earn its place.
Topic hubs
AI agent security topic hub
AI Agent Security Articles
Clawmont articles on AI agent security: runtime protection, tools, posture management, identity, monitoring, data exfiltration, testing, and audit evidence.
MCP security topic hub
MCP Security Articles
Clawmont articles on MCP security for AI agents: MCP scanners, gateways, authorization, server hardening, tool poisoning, runtime tool-call checks, and audit evidence.
Agentic security topic hub
Agentic Security Articles
Clawmont articles on agentic security: agentic AI security frameworks, runtime controls, memory risk, supply chain risk, ransomware, governance, and audit evidence.
Runtime security topic hub
AI Agent Runtime Security Articles
Clawmont articles on AI agent runtime security: pre-tool checks, post-tool response screening, runtime policy enforcement, MCP controls, monitoring, and local OpenClaw audit evidence.
Guides
Evergreen references for the highest-intent AI agent security questions.
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.
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.
AI Agent Memory Poisoning: Persistent Memory Risk and Runtime Defense
Poisoned memory persists: instructions planted in an agent's stored context re-enter every later session. Why retrieval is the choke point, and the provenance, audit, and rollback controls that limit the damage.
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.
NSA MCP Security Guidance: What It Means for AI Agents
NSA, CSA, IETF, and official MCP guidance translated into runtime controls for AI agents: trust boundaries, identity, tool validation, execution checks, 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.
What 22 Claude Code Security Advisories Teach Us About AI Agent Security
Claude Code advisories and the July 2026 reverse-shell case show the same pattern: the dangerous decision often happens at runtime, after static review says the project looks clean.