HackAgent is an open-source Python SDK and CLI designed for security researchers, developers, and AI safety practitioners. It addresses the growing need to evaluate and harden AI agents against novel security threats like prompt injection, jailbreaking, and goal hijacking, which traditional tools cannot detect. By automating vulnerability testing with research-backed attack techniques, HackAgent helps identify and mitigate risks, enhancing the overall security posture of AI systems.

How It Works

HackAgent employs a modular pipeline architecture to test AI agent robustness. The core components include an Attack Engine that orchestrates various adversarial techniques (e.g., AdvPrefix, AutoDAN-Turbo, PAIR), a Generator LLM responsible for creating malicious prompts, and a Judge LLM that assesses whether attacks successfully bypass safety measures. The system tests a Target Agent, leveraging pre-built or custom datasets, and supports multiple AI agent frameworks. This LLM-driven, modular approach allows for flexible and comprehensive security evaluations.

Quick Start & Requirements

• Install: python3 -m venv .venv && source .venv/bin/activate && pip install hackagent
• Prerequisites: Python 3.10+. No API key is required for local operation.
• Links: Docs, App

Highlighted Details

• Supports a wide array of advanced attack techniques including AdvPrefix, AutoDAN-Turbo, PAIR, TAP, FlipAttack, BoN, h4rm3l, CipherChat, PAP, and Static Template.
• Features an interactive Text User Interface (TUI) providing real-time attack progress visualization.
• Offers both local reporting (SQLite database with a built-in dashboard) and optional cloud reporting via the HackAgent platform.

Maintenance & Community

Community discussions are available via email at ais@ai4i.it. Contribution guidelines are detailed in CONTRIBUTING.md, and a CODE_OF_CONDUCT.md is provided.

Licensing & Compatibility

Licensed under the Apache-2.0 license. This license is permissive and generally allows for commercial use and integration into closed-source projects.

Limitations & Caveats

HackAgent is strictly intended for authorized security testing and AI safety research. The project explicitly prohibits testing systems without permission, malicious exploitation, or violating terms of service, with authors disclaiming responsibility for any misuse.