Agent Security

Cloudflare's x402 and paid MCP tooling make agentic payments real, but payment proof is not runtime permission. This guide explains spend authorization, consent tiers, and audit requirements for paid tool calls.

MCP is making agentic ERP integration easier, but security now depends on runtime authorization at the tool-call layer. Learn how to model scoped permissions, trust levels, and audit evidence for finance, HR, procurement, and payroll workflows.

MCP risk is not frozen at build time. This article explains how to vet third-party MCP servers, treat manifests as security boundaries, enforce runtime authorization, and preserve incident-grade audit evidence.

CVE-2026-49257 in mcp-pinot shows why network-reachable MCP database servers must fail closed: secure startup, endpoint authentication, and per-tool runtime authorization are all mandatory. This guide breaks down the confused-deputy pattern, risk-tiered tool policy for read vs schema/admin operations, and the audit model needed for real incident forensics.

Specs and PRDs make coding agents more accurate, but not inherently safe. This guide explains how to secure MCP-enabled coding and workflow agents with short-lived delegated access, runtime policy decisions, and auditable zero standing permissions.

Agent traces are observability signals, not accountability evidence. This guide defines a causal, replayable authorization commit log for MCP tool calls so teams can prove who delegated access, why policy allowed an action, and what happened next.

Subagents are delegated actors, not implementation details. This guide explains how to design MCP permission delegation, OAuth token brokering, approval routing, and audit trails to avoid silent stalls and privilege expansion.

OpenAPI-to-MCP gateways can instantly expose REST endpoints as agent tools, but endpoint filtering alone does not enforce least privilege. This guide shows how to classify generated tools by risk, enforce runtime policy per delegator and intent, keep credentials server-side, and log auditable MCP decisions.

Text refusal and tool behavior can diverge in coding agents. This article explains why runtime, action-time authorization is the real security boundary for Codex, Claude Code, Cursor, and MCP tool calls.

Hermes Agent's Blank Slate direction shows why teams are moving from broad default tool access to zero standing permissions with config pinning and runtime authorization. This guide explains the safest local defaults, risk-tiered tool access, and practical temporary grant patterns for web, browser, terminal, MCP, memory, and delegation.

DIDs, verifiable credentials, and AI control towers are foundational for agent governance, but they still do not decide whether a specific agent action is allowed right now. This article explains the runtime authorization model enterprises need for delegated AI execution.

Prompt injection becomes a security incident when untrusted content is promoted across authority boundaries into actions. This article shows how to enforce RAG and MCP promotion gates with runtime authorization outside the model.

It had every permission it needed and a ticket telling it exactly what to do. Blocked once, it reworded the request to fool the check. Blocked again, it asked me to switch the check off. This is the call-by-call trace of why nothing left Linear — and the design decision that made "reword it until it's allowed" a dead end.

The LiteLLM CVE-2026-42271 and Starlette BadHost CVE-2026-48710 chain turned authenticated command injection into unauthenticated RCE. The deeper lesson: AI gateways hold model credentials, route sensitive traffic, and expose MCP utility endpoints — and need action-time authorization, not flat API keys.

Sandboxing a coding agent isolates it from the host—but the real blast radius is the credentials it holds. GitHub tokens, cloud keys, MCP connections, and CI/CD access define what an agent can actually do. Here's the runtime permission model that closes the gap.

Atlassian Rovo's MCP server makes a precise security tradeoff visible: OAuth 2.1 handles identity and consent; API tokens handle non-interactive automation. Neither governs what agents can actually do at tool-call time. Here is what that gap looks like in practice.

Coding agents are operational actors, not just assistants. This guide presents a practical trust-level taxonomy for agent commands and MCP tools, explains why human approval prompts degrade at scale, and shows how runtime authorization policy enforces trust levels without relying on click fatigue.

Treating AI agents like service accounts is a useful starting point — but it fails at runtime. Here's why scoped tokens are necessary but not sufficient, and how runtime authorization fills the gap.

The Claude Code MCP OAuth token theft chain is an authorization failure, not just a credential leak. OAuth got the agent connected, but it never constrained which tool calls remained valid after the routing layer was tampered with — and that is the gap runtime policy enforcement must close.

The IETF SD Agent draft and Microsoft Entra Agent ID are turning agent identity into real infrastructure. But a verified Agent Card or sponsored enterprise identity still doesn't answer whether an agent may call a specific MCP tool right now — that requires runtime authorization.

Microsoft Entra Agent ID and SD-JWT agent identity solve registration, governance, and authentication — but they don't decide whether a specific MCP tool call is permissible right now. This article explains the gap and the runtime authorization architecture needed to close it.

In April 2026, the NSA published 'Careful Adoption of Agentic AI Services' — the first intelligence-community advisory specifically targeting AI agent authorization failures. Here is what it actually demands and why most engineering teams are not close to meeting it.

If you already run OPA, AI agents don't require a new policy engine — they require a richer input schema, ephemeral identity, and enforcement at every layer. Learn how to evolve your OPA setup for delegated, multi-hop agentic authorization with Zero Standing Permissions, production-grade Rego, and OPAL-backed real-time enforcement.

Zero Standing Privileges (ZSP) means no identity holds usable access between tasks. This article explains how ZSP differs from least privilege, how to implement it with ephemeral credentials and runtime policy enforcement, and why AI agents running on MCP make standing access a new category of operational risk.

OAuth 2.1 is the right foundation for MCP security, but most implementations stop one layer too early. This guide covers every spec-required piece: protected resource metadata, authorization server discovery, PKCE, dynamic client registration, resource indicators, and where fine-grained authorization picks up where OAuth ends.

AI agents acting on behalf of users need more than authentication — they need governance. This article covers the Permit.io agentic identity model, policy-as-code lifecycle, MCP Gateway enforcement, zero standing credentials, Guardian Agents, and what an audit trail must contain to be meaningful.

Coding agents execute code, run commands, and call APIs — not just generate text. This guide covers the real security risks, why authorization must happen at the tool-call level, and how Permit.io and the Permit MCP Gateway enforce least-privilege access for agentic workflows.

AI agents acting across tools, APIs, and multi-agent pipelines raise hard questions about identity, authentication, and authorization. This article covers agentic identity (delegating human + workflow context + intent), agent interrogation, SPIFFE, OAuth token exchange, least privilege at runtime, and what audit actually means for agentic systems.

AI agents break the traditional least-privilege model. This article explains why, defines agentic identity (delegating human + workflow context + declared intent), and shows how Permit.io enforces zero standing privileges through gateway-vaulted credentials, the PDP, MCP Gateway, and downscoped delegation chains.