How to Validate URLs Before an AI Agent Opens Them
Validate URLs before AI agents fetch or open them — protocol restrictions, domain allowlists, redirect handling, and integration with action policies.
Blog / AI Agents
AI Agents
Agent permissions, tool security, least privilege, and controlling excessive agency in autonomous workflows.
URL-Based Data Exfiltration in AI Agents
Prevent URL-based data exfiltration in AI agents — destination controls, sensitive-data detection, least privilege, and action policy gates.
URL Allowlist vs Domain Allowlist for AI Agents
Compare URL allowlists and domain allowlists for AI agents — exact paths, subdomains, schemes, redirects, and practical policy trade-offs.
How to Threat Model an AI Agent
Threat model an AI agent — tools, permissions, external content, credentials, high-impact actions, human approval, and runtime monitoring.
Human-in-the-Loop Approval for AI Agent Actions
When should AI agent actions require human approval? Learn approval boundaries for financial, destructive, and external operations without approval fatigue.
Allow, Review or Block: Designing AI Agent Policies
Design practical AI agent policies with allow, review, and block decisions. Deterministic rules, policy evaluation order, and when to escalate.
How to Prevent AI Agents from Sending Sensitive Data
Prevent AI agent data exfiltration with least privilege, destination validation, secret scanning, and action policies before external sends.
Securing AI Agents That Send Email
Secure AI email agents with recipient validation, draft vs send permissions, approval workflows, and sensitive-data checks before outbound mail.
Securing AI Agents with Database Access
Secure database access for AI agents — read vs write permissions, scoped credentials, destructive-operation approval, and audit metadata.
Securing AI Agents That Browse the Web
Secure web-browsing agents against untrusted page content, indirect prompt injection, URL validation, and unsafe tool chains.
AI Agent Tool Allowlist vs Blocklist
Compare allowlist and blocklist strategies for AI agent tools — when each fits, and how to combine them for read-only, external, and destructive operations.
How to Design Safe Destructive Actions for AI Agents
Design safe destructive agent actions — deletion, cancellation, revocation — with preview, confirmation, authorization, and audit events.
Runtime Security for AI Agents
Runtime security for AI agents — policy evaluation at tool request time, allow/review/block decisions, and integration before execution.
AI Agent Security Checklist
A production checklist for AI agent security — identity, credentials, tools, permissions, untrusted content, external communication, and incident response.
What Is MCP Security?
MCP (Model Context Protocol) security for AI agents — tool exposure, authentication boundaries, untrusted content, and application-level policy enforcement.
How to Secure MCP Servers Used by AI Agents
Secure MCP servers with authentication, least privilege, exposed tool review, transport boundaries, secrets handling, and safe logging.
MCP Tool Permissions: Security Best Practices
MCP tool permission best practices — scope exposure, read vs write tools, and application policy layers that complement protocol-level access.
How to Validate AI Tool Calls Before Execution
Validate AI tool calls with schema checks, permission evaluation, secret scanning, and policy decisions before allow, review, or block execution.
Tool Output Injection in AI Agents
Tool output injection treats tool results as potentially hostile instructions. Learn detection, sanitization, and policy gates before re-prompting.
Source-to-Sink Security for AI Agents
Source-to-sink security prevents untrusted sources (web, email, documents) from driving high-impact sinks (send, delete, purchase) without policy gates.
Securing Long-Running AI Agents
Long-running agent security — stale permissions, accumulated context, repeated tool calls, budgets, re-authorization, time limits, and escalation.
AI Agent Runtime Monitoring: What Should You Log?
Privacy-safe AI agent monitoring — log tool requests, policy decisions, latency, and outcomes without storing secrets or raw sensitive prompts.
MCP Security Checklist for Production
A production MCP security checklist — server hardening, tool exposure, authentication, permissions, untrusted content, logging, and incident response.
What Is AI Agent Security?
AI agent security covers tool permissions, action policies, untrusted content ingestion, and human oversight for autonomous LLM workflows.
How to Secure AI Agents Before They Use Tools
Secure tool-calling agents with permission boundaries, action policies, secret scanning, and approval workflows before tools execute.
AI Agent Permissions: A Developer's Guide
Design agent permission models — scoped tools, read vs write actions, destructive operation controls, and policy-based decisions.
Excessive Agency in LLM Applications Explained
Excessive agency happens when agents have too much functionality, permission, or autonomy. Learn risks and how to apply least privilege.
How to Apply Least Privilege to AI Agents
Apply least privilege to AI agents — minimal tool sets, scoped credentials, approval for high-impact actions, and continuous policy review.