deepteam

What it is and what it does

DeepTeam is a red teaming framework built on top of DeepEval that lets you test LLM systems for security and safety vulnerabilities by simulating attacks locally on your machine. It covers a broad attack surface: data privacy (PII leakage, prompt leakage), responsible AI (bias, toxicity, fairness), security (SQL injection, SSRF, authorization bypasses), safety (illegal activity, self-harm), business risks (misinformation, IP violations), and agentic-specific threats (goal theft, recursive hijacking, tool abuse). Each vulnerability is tested using LLM-as-a-Judge metrics that produce binary pass/fail scores with reasoning.

The framework runs on Python 3.9–3.13 and depends on common packages like aiohttp, openai, grpcio, and requests. It offers both pre-built vulnerability checks and the ability to define custom vulnerabilities in code. Results can be managed through the optional Confident AI platform for risk tracking and team reporting, though the core framework runs entirely offline.

Use it for:

• Test a chatbot or AI agent for prompt injection and jailbreak vulnerabilities before deployment.
• Audit a RAG pipeline for PII leakage and prompt leakage in retrieved context.
• Verify that an LLM-powered system refuses illegal activity, self-harm requests, and harmful content.
• Check multi-turn agent workflows for goal theft, recursive hijacking, and tool orchestration abuse.
• Validate that an AI system exhibits fair treatment across demographic groups and does not exhibit bias.
• Scan for authorization bypasses (BOLA, BFLA, RBAC) in LLM-powered APIs and tool-calling agents.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

DeepTeam is an open-source red teaming framework that simulates attacks on LLM systems to uncover vulnerabilities like bias, PII leakage, and prompt injection, and provides guardrails to mitigate them in production.

Yes. DeepTeam is actively maintained, has no known vulnerabilities, runs with low install friction, and addresses a critical gap in LLM security testing. It is permissively licensed and supports current Python versions. Install it if you are building or deploying LLM systems and need systematic vulnerability assessment before production.

Install

deepteam on PyPI

pip install deepteam

uv add deepteam

poetry add deepteam

Installing deepteam

Before you install

Low install friction with a pure-Python wheel distribution. Active maintenance with a recent release (2 days old) and 2445 repository stars. Depends on 8 runtime packages including aiohttp, openai, and grpcio, all of which are widely used.

License in practice

Apache-2.0 permissive license allows commercial and private use with minimal restrictions, requiring only attribution and disclosure of modifications.

Quickstart

pip install deepteam

from deepteam import RedTeamer
from openai import OpenAI

client = OpenAI()
red_teamer = RedTeamer(client=client)
results = red_teamer.run_vulnerability_scan()

Requires an LLM API key (e.g., OpenAI) to run attacks; local evaluation uses an LLM-as-a-Judge pattern, so you need access to a model for both attack generation and scoring.

Verify before relying

• Whether the 50+ vulnerabilities and 20+ attack methods are all functional in version 1.0.9 or if some are still in development.
• Performance characteristics and typical runtime for scanning a single LLM endpoint or agent.
• Whether guardrails can be deployed as middleware or require code changes to the target LLM system.

Package facts

Evidence: deepteam-1.0.9-py3-none-any.whl

Tags

More Testing packages

Further reading