ai-ml-security

Assess security threats across AI/ML supply chains, adversarial examples, and data privacy.

5|2|Updated May 16, 2026
One-click install
npx skills add https://github.com/DorianGallo/hack-skills-local --skill ai-ml-security-doriangallo
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: ai-ml-security
Source: https://github.com/DorianGallo/hack-skills-local/tree/main/skills/ai-ml-security
Command: npx skills add https://github.com/DorianGallo/hack-skills-local --skill ai-ml-security-doriangallo

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill helps you assess and mitigate real-world security risks across AI/ML pipelines, including model supply chain compromise, adversarial manipulation, data poisoning, model extraction, and privacy leakage from trained models.

Core Features & Use Cases

  • Model Supply Chain Attacks: Diagnose and reduce risks from malicious model files (e.g., pickle-based RCE), unsafe model formats, and poisoned or backdoored artifacts from external repositories.
  • Adversarial Examples & Robustness Testing: Understand and operationalize common attacker strategies (FGSM, PGD, C&W) and physical-world adversarial patterns.
  • Privacy & Extraction Threats: Evaluate membership inference, model inversion, and gradient leakage risks, and understand query-based model stealing/extraction.

Quick Start

Use the ai-ml-security skill to build an end-to-end threat checklist for your model loading, training pipeline, inference API, and privacy posture.

Frequently Asked Questions about ai-ml-security

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I secure my ML model supply chain against malicious model files?

Secure ML model supply chains by diagnosing risks from malicious pickle-based RCE, unsafe model formats, and poisoned artifacts from external repositories, applying safe model formats and trust boundaries to mitigate compromise during model loading and deployment pipelines.

What are adversarial examples and how do attack strategies like FGSM and PGD work?

Adversarial examples are manipulated inputs designed to fool ML models. Attackers operationalize strategies like FGSM, PGD, and C&W to exploit model robustness vulnerabilities, including physical-world adversarial patterns that compromise inference API integrity.

How can I prevent model extraction and membership inference attacks on my inference API?

Prevent model extraction and membership inference attacks by evaluating query-based model stealing risks, gradient leakage, and model inversion vulnerabilities across your inference APIs, implementing structured privacy threat mitigations to protect training data.

Does this cover LLM agent security and autonomous tool use vulnerabilities?

LLM agent security is covered by assessing autonomous tool use and memorization risks in LLM-adjacent workflows, cross-linking to related LLM and deserialization skills for deeper attack execution analysis and threat mitigation.

What's the best way to build an end-to-end threat checklist for AI/ML pipelines?

Build an end-to-end threat checklist by assessing security threats across model loading, training, federated learning, and inference API setups, ensuring structured threat coverage and actionable mitigations for data privacy leakage and supply chain compromise.