DeepTeam (by Confident AI, makers of DeepEval) is an open-source LLM red teaming framework with OWASP-mapped vulnerabilities. ai-blackteam is independent and ships a larger curated attack corpus with broader compliance mapping.

At a glance

ai-blackteamDeepTeam
LicenseMITApache 2.0
Backed byIndependentConfident AI
Installpip install ai-blackteampip install deepteam
Curated attacks1,020Vulnerability + attack modules
Adaptive attacksCrescendo, TAP, PAIRMulti-turn attack methods
Providers17 (16 vendors + generic HTTP)Pluggable model callbacks
Compliance mapsOWASP LLM + Agentic, MITRE ATLAS, MLCommons, NIST, EU AI ActOWASP LLM + Agentic Top 10
EcosystemStandalone, exports to Promptfoo/garak/SARIFPairs with DeepEval evaluation suite

When DeepTeam fits

  • You already use DeepEval and want red teaming in the same ecosystem
  • You want a compact, modular framework with custom vulnerability definitions

When ai-blackteam fits

  • You want a larger curated attack corpus (1,020 techniques) ready to run
  • You want mapping to five compliance frameworks for audit reporting, not just OWASP
  • You want SARIF export for the GitHub Security tab and a generic HTTP provider for your own endpoints
  • You want a vendor-neutral tool that isn’t tied to a specific evaluation product

Summary

Both map to the OWASP Agentic Top 10 2026. ai-blackteam differentiates on corpus size, multi-framework compliance mapping, and export breadth (Promptfoo, garak, SARIF).