DeepTeam is a framework to red team LLMs and AI agents.
.
penguine1b671ba37682
diff --git a/docs/guides/guide-agentic-ai-red-teaming.mdx b/docs/guides/guide-agentic-ai-red-teaming.mdxindex d0647f5c..ffbb60e2 100644--- a/docs/guides/guide-agentic-ai-red-teaming.mdx+++ b/docs/guides/guide-agentic-ai-red-teaming.mdx@@ -15,6 +15,16 @@ This guide explains agentic AI red teaming with [DeepTeam](https://github.com/coThis guide focuses on agentic red teaming for tool-calling AI agents. For RAG-based agents, use the [agentic RAG guide](/guides/guide-red-teaming-agentic-rag). For conversational agents, use the [conversational agents guide](/guides/guide-red-teaming-conversational-agents).:::+## What is Agentic AI Red Teaming?++Agentic AI red teaming is the practice of adversarially testing AI agents that go beyond generating text — agents that **reason, plan, call tools, delegate to sub-agents, and take real-world actions**. Unlike traditional prompt-level red teaming, agentic red teaming targets the full autonomous pipeline:++- **Reasoning and planning** — Can an attacker redirect the agent's goals or make it "forget" constraints from its system prompt?+- **Tool calling and actions** — Can the agent be tricked into executing unauthorized commands, chaining tools in dangerous sequences, or passing manipulated parameters?+- **Inter-agent communication** — In multi-agent systems, can adversarial payloads traverse trust boundaries between agents?++A compromised AI agent won't throw errors. It will delete files, leak credentials, or make unauthorized API calls — all while appearing to function normally. That's what makes agentic red teaming essential.+## The AI Agent Attack SurfaceTraditional LLM applications have one attack surface: the prompt-response interface. AI agents have three, and each one can be compromised independently.diff --git a/docs/guides/guide-red-teaming-agentic-rag.mdx b/docs/guides/guide-red-teaming-agentic-rag.mdxindex c5b667eb..9dea9659 100644--- a/docs/guides/guide-red-teaming-agentic-rag.mdx+++ b/docs/guides/guide-red-teaming-agentic-rag.mdx@@ -15,6 +15,17 @@ This guide walks through the methodology for red teaming agentic RAG and RAG-basFor conversational AI agents that maintain chat history across turns, use the [conversational agents guide](/guides/guide-red-teaming-conversational-agents). For AI agents with tool-calling capabilities, use the [AI agents guide](/guides/guide-agentic-ai-red-teaming).:::+## What is Agentic RAG Red Teaming?++Agentic RAG red teaming is the practice of adversarially testing AI agents backed by **retrieval-augmented generation (RAG)** — systems where a retriever pulls context from a knowledge base and an agent generates responses grounded in that context. It targets the unique failure modes that emerge when retrieval and generation interact:++- **Retrieval poisoning** — Can adversarial or sensitive documents in the knowledge base override the agent's instructions or leak through its responses?+- **Access control bypass** — Can a user retrieve documents they shouldn't have access to by crafting the right query?+- **Grounding failures** — Does the agent fabricate policies, citations, or procedures that aren't supported by any retrieved document?+- **Instruction override** — Can an attacker break the agent out of its assigned role through prompt injection?++These failures are invisible on the surface — no errors, no stack traces, just a confident response that shouldn't exist. Agentic RAG red teaming is how you find them before users do.+## Think in Threat Models, Not ChecklistsThe biggest mistake teams make is picking vulnerabilities from a list without thinking about their system first. A customer-facing FAQ agent and an internal HR agent over sensitive employee records have fundamentally different risk profiles—even though they're both agentic RAG pipelines.diff --git a/docs/guides/guide-red-teaming-conversational-agents.mdx b/docs/guides/guide-red-teaming-conversational-agents.mdxindex f55e47b2..47b5dcc8 100644--- a/docs/guides/guide-red-teaming-conversational-agents.mdx+++ b/docs/guides/guide-red-teaming-conversational-agents.mdx@@ -15,6 +15,16 @@ This guide explains how to red team a conversational AI agent using [DeepTeam](hThis guide focuses on multi-turn attack techniques for conversational agents. If your AI agent handles requests in a single generation without maintaining conversation state, use the [agentic RAG guide](/guides/guide-red-teaming-agentic-rag) instead. For AI agents with tool-calling capabilities, use the [AI agents guide](/guides/guide-agentic-ai-red-teaming).:::+## What is Conversational Agent Red Teaming?++Conversational agent red teaming is the practice of adversarially testing AI agents that **maintain state across multiple turns of dialogue**. Unlike single-turn red teaming, it targets vulnerabilities that only emerge over the course of a conversation:++- **Consistency pressure** — Once an agent commits to a tone or line of reasoning, attackers can gradually steer it into territory it would refuse from a cold start.+- **Information leakage across turns** — Each response reveals something about the agent's boundaries, giving attackers a map of its defenses.+- **Context window saturation** — As conversation history grows, safety instructions get pushed out of the model's attention window.++An agent that's perfectly safe on any single message can still fail catastrophically across a five-turn conversation. Conversational red teaming tests the conversation as a unit, not individual turns.+## Why Multi-Turn Attacks WorkA single-turn attack gives you one shot — the prompt either breaks through or it doesn't. Multi-turn attacks are fundamentally more powerful because they exploit three properties of conversation that don't exist in single-turn interactions.