BlackIce Introduced as Container-Based Red Teaming Toolkit for AI Security Testing
Databricks has announced the release of BlackIce, an open-source toolkit introduced at CAMLIS Red 2025. This containerized solution integrates 14 AI security tools into a single environment, streamlining the setup and reducing dependency conflicts in AI…
Databricks has announced the release of BlackIce, an open-source toolkit introduced at CAMLIS Red 2025. This containerized solution integrates 14 AI security tools into a single environment, streamlining the setup and reducing dependency conflicts in AI red teaming processes.
BlackIce is designed to overcome several obstacles in security testing:
Complex individual tool setups are consolidated into one streamlined process. Dependency conflicts are minimized by maintaining separate runtime environments within the toolkit. Testing flexibility is enhanced by supporting multiple Python interpreters. Newcomers are supported with pre-configured tools that simplify AI security testing.
The toolkit takes inspiration from Kali Linux’s approach to penetration testing, packaging tools into a unified Docker container.
Core Capabilities and Integrated Tools
BlackIce includes 14 open-source tools covering Responsible AI, security testing, and adversarial machine learning. Notable tools include:
LM Eval Harness Promptfoo Microsoft’s PyRIT CyberArk’s Fuzzy AI
These tools facilitate comprehensive vulnerability assessments across various attack vectors.
Feature Description
Containerized toolkit Version-pinned Docker image for reproducible environments.
Databricks has announced the release of BlackIce, an open-source toolkit introduced at CAMLIS Red 2025.
14 integrated AI security tools Comprehensive toolset for AI security and adversarial testing.
Unified CLI Single command-line interface for all tools.
Static and dynamic tool modes CLI for static tools, Python customization for dynamic tools.
Isolated environments for static tools Separate virtual environments to avoid dependency conflicts.
Global environment for dynamic tools Centralized Python environment with shared requirements.
Databricks-native integration Direct communication with Databricks Model Serving and workspaces.
Framework-aligned coverage Aligned with MITRE ATLAS and DASF standards.
Supply chain and artifact scanning Detects unsafe AI artifacts and malicious files.
Cloud-friendly deployment Optimized for Databricks Container Services.
BlackIce aligns with the MITRE ATLAS and Databricks AI Security Framework (DASF), addressing vulnerabilities such as:
Prompt injection LLM jailbreaks Data leakage Hallucination detection Supply chain artifact scanning
This alignment ensures comprehensive threat modeling and identification of potential weaknesses.
BlackIce can be accessed on Databricks’ Docker Hub and deployed using standard commands. It integrates with Databricks Model Serving endpoints, facilitating seamless compatibility with existing workflows.
The GitHub repository provides build documentation for customization and version management.
To access BlackIce, use docker pull databricksruntime/blackice:17.3-LTS and configure environments via Databricks Container Services. A demo notebook and the GitHub repository offer implementation examples.
Based on reporting by GBHackers.