Arcade's $60M Round Makes Agent Permissions Infrastructure
Arcade's funding round is a signal that agent authorization, MCP tool permissions, and auditability are becoming core AI infrastructure.
MCP Security
Tool poisoning, prompt injection through tool output, and config-level attacks on the Model Context Protocol — how they work and how to scan for them.
The Model Context Protocol gives an agent a typed list of tools it can call, each described in natural language. That description is part of the model's prompt, which makes an MCP server a write channel into the agent's instructions — a hostile or compromised server can poison a tool description, smuggle directives through tool output, or hide metadata a human reviewing the config never sees.
These posts cover the documented attack classes, the standards work responding to them, and the practical checks. SkillSafe's MCP scanner reads a config and flags poisoning and exfiltration patterns before the agent ever connects.
12 articles in this guide
Agentjacking Turns MCP Tool Output Into an Execution Path
Tenet's Agentjacking research shows how fake Sentry errors can hijack coding agents through MCP tool output. The defense starts with treating tool data as untrusted.
IETF MCP Draft Turns Tool Safety Into a Standards Problem
A new MCP security Internet-Draft ties public SSRF reports, tool permission drift, and protocol pivoting into one operational warning for agent teams.
Runaway Agent AWS Bill Shows Tool Budgets Are Permissions
A runaway AI agent's DN42 scanning attempt turned into a $6,531 AWS bill. The lesson for agent tools is simple: budgets, scope, and blast radius are permissions.
OpenClaw Shows Why Agents Need a Bill of Materials
Renewed discussion of OpenClaw's local-agent takeover risk shows why teams need an Agent Bill of Materials for skills, plugins, MCP servers, and connectors.
Cisco Cloud Control Makes Agent Tool Governance Mainstream
Cisco's Cloud Control launch puts AI agents, MCP connectors, and third-party tool marketplaces inside critical infrastructure operations.
VIPER-MCP Shows Agent Tools Need Taint Scanning
VIPER-MCP found 106 confirmed zero-days across nearly 40,000 MCP server repos. Agent tool security now needs code-level taint analysis, not just trust prompts.
Microsoft ACS: Agent Controls Move Into the Runtime
Microsoft's Agent Control Specification gives agent teams a portable runtime policy layer for tool calls, approvals, and audit evidence.
NSA MCP Guidance: Inventory Agent Tools Before They Drift
NSA's MCP security guidance turns the agent tooling debate into an operational checklist: inventory servers, verify tool changes, and scan before trust drifts.
MCP RCE Debate: Treat Agent Plugins Like Executable Code
The public debate around MCP remote code execution risk shows a hard lesson for AI agents: plugins, connectors, and skills need supply-chain controls.
MCP Tool Poisoning: How Hidden Metadata Hijacks AI Agents
MCP tool descriptions are visible to your AI agent but hidden from you. Attackers embed instructions that hijack agent behavior and steal credentials.
You're Using Claude Code Skills. Do You Know What's in Them?
Claude Code skills can read files, run commands, and access credentials. What the skill ecosystem gets wrong about security — and how to protect yourself.