Penetration testing for AI systems.

Framework-aligned risk assessments straight from Python — OWASP Top 10, NIST AI RMF, MITRE ATLAS, and more. Surface jailbreaks, vulnerabilities, and agentic risks before they ship.

main.py

from deepteam import red_team
from deepteam.test_case import RTTurn
from deepteam.frameworks import OWASPTop10
 
def model_callback(prompt: str, turns: List[RTTurn]= None) -> RTTurn:
 return your_llm_app(prompt, turns)
 
red_team(model_callback=model_callback, framework=OWASPTop10())

terminal

$python main.py

Run Red Team

LLM as your adversary.

DeepTeam uses LLMs to simulate adversarial attacks across every vulnerability and attack vectors so you find the failures before real attackers do.

Industry-standrd frameworks for your trust & safety requirements.

OWASP Top 10 for LLMs, MITRE ATLAS, NIST AI RMF, EU AI Act — plus safety benchmark datasets like BeaverTails and Aegis. Every major taxonomy plugged into a single framework.

Any model. Any framework. Any pipeline.

DeepTeam plugs into the LLM providers you already use, the safety frameworks your auditors require, and the CI/CD runners your team already trusts — continuous red teaming with zero glue code.

Built by amazing humans.

Nothing would be possible without our community of amazing contributors, thank you!

A-Vamshi — 577 contributionspenguine-ip — 345 contributionssid-murali — 105 contributionskritinv — 28 contributionssergeyklay — 11 contributionsAmineDjeghri — 9 contributionsabhinavbom — 5 contributionssayan5069 — 4 contributionsArun-Niranjan — 2 contributionskarankulshrestha — 2 contributionsSKM162 — 2 contributionsspike-spiegel-21 — 2 contributionsabhinav-wingback — 1 contributionexecuteautomation — 1 contributionfrankentini — 1 contributionkarthick965938 — 1 contributionrauschecker — 1 contributionSai-Suraj-27 — 1 contributionSidhaarth-Murali — 1 contributionsubbyte — 1 contributiontanayvaswani — 1 contributiontrevor-inflection — 1 contribution

Ah yes, FAQs.

This is the CTA :)

Get Started