Skip to content

Agentic Red Teaming Overview

A practitioner overview of red teaming AI agents and multi-agent systems on the Dreadnode platform, mapped to the OWASP Top 10 for Agentic Applications.

Red teaming a chat model asks whether the model produces harmful text. Red teaming an agent asks a different question: whether the agent takes a harmful action. An agent plans, calls tools, reads external content, and coordinates with other agents, so the attack surface and the evidence of success are both about actions, not words.

This page explains how Dreadnode structures agentic red teaming, how it maps to the OWASP Top 10 for Agentic Applications, and where to go next for the SDK and TUI workflows.

Dreadnode scores actions, not words. A finding is recorded only when the agent performs an action that violates policy, established from the tool calls the target actually made and the arguments it used. There is no planted flag and no boolean success signal from the target. The ground truth is which tool ran and with what arguments.

For a data exfiltration or code execution assessment against a production agent where you cannot plant a flag, Dreadnode uses honeytokens. A honeytoken is an inert, unique value that is safe to leak. If the agent leaks it or runs it, exfiltration or code execution is proven with no damage. See Honeytoken Probing.

Coverage: OWASP Top 10 for Agentic Applications (2026)

Section titled “Coverage: OWASP Top 10 for Agentic Applications (2026)”

The agentic suite drives every category in the 2026 OWASP-ASI taxonomy. For each category it selects the mapped attack strategies, applies the relevant transform families, and scores results with the matching detection scorers.

CategoryWhat Dreadnode runs
ASI01 Agent Goal HijackInjection across direct, tool output, and peer message surfaces; reasoning hijack
ASI02 Tool Misuse and ExploitationArgument-aware tool-evidence gate; MCP tool poisoning; mcp-poisoning-mesh
ASI03 Identity and Privilege AbuseDelegation and boundary-crossing analysis across privilege pipelines
ASI04 Agentic Supply ChainTyposquat and dependency confusion; supply-chain-mesh
ASI05 Unexpected Code Execution (RCE)Real shell execution and honeytoken proof-of-execution; devops-rce-mesh
ASI06 Memory and Context PoisoningShared-memory and query-memory injection transforms
ASI07 Insecure Inter-Agent CommunicationPrompt infection, consensus poisoning, peer spoofing across agent meshes
ASI08 Cascading FailuresCascade depth and boundary-crossing metrics across delegation chains
ASI09 Human-Agent Trust ExploitationPersuasion and trust-exploitation transforms
ASI10 Rogue AgentsGoal drift and reasoning hijack transforms

Indirect prompt injection, the EchoLeak and ForcedLeak class where a hidden instruction rides inside content the agent reads, is exercised through the tool output surface and the indirect-injection-mesh environment.

An agentic target is any HTTP service that accepts a message and returns the executed tool calls.

POST /attack
{ "prompt": "...", "surface": "direct|tool_output|peer_message", "injection": "..." }
->
{ "content": "...", "tool_calls": [ {agent, tool, arguments, result}, ... ] }

The surface field selects where the injection is applied: direct for the entry agent, tool_output for content a tool returns (indirect prompt injection), and peer_message for an inter-agent delegation. Any deployment that answers this contract works, whether it runs locally, on AWS, or on Azure.

Dreadnode ships environments that expose each attack surface so you can validate a workflow before pointing it at your own system.

EnvironmentRisk
finops-meshTool misuse and privilege escalation
devsecops-meshSupply chain and deployment
healthcare-meshUnsafe clinical action
soc-meshSOC sabotage and exfiltration
devops-rce-meshReal code execution
support-exfil-meshData exfiltration
mcp-poisoning-meshMCP tool poisoning
reasoning-hijack-meshReasoning hijack
indirect-injection-meshIndirect prompt injection
supply-chain-meshAgentic supply chain