T-2002ObservedActive

Indirect Prompt Injection

Embed malicious instructions in external data sources consumed by the agent via RAG or tool responses

Tactic

Initial Access · Stage 2

Gain control over agent behavior through prompt manipulation or input exploitation

Attack class

RAG-POISON

Injecting malicious content into retrieval-augmented generation data sources

Evidence grade
Observed

Confirmed in real-world production systems or internet-wide exposure assessments.

DVAA validation

RAGBot

Reproductions in Damn Vulnerable AI Agent, the OpenA2A intentionally-broken agent for kill-chain validation.

Honeypot

AgentPwn coverage

Live
prompt-injectionagentpwn.com/learn ↗

An AgentPwn trap page produces a payload tagged with this technique class. Following the AgentPwn taxonomy of trap pages shows what an agent encounters.

Every AgentPwn page is an indirect-injection surface delivered through consumed web content.

Detect

Detection · HackMyAgent

Live5 live · 0 queued
PROMPT-001RAG-001RAG-002RAG-003RAG-004
npx hackmyagent secure --ciLive = implemented in hackmyagent; queued = declared
Defend

Defense · OASB controls

Live5 live · 0 queued
Live = documented at oasb.ai; queued = declared
Reference

How to cite

AI Agent Threat Matrix T-2002 (Indirect Prompt Injection). OpenA2A, 2026. https://threats.opena2a.org/techniques/T-2002