What’s Inside Tencent’s AI Security Attack Matrix? A Minefield Guide

What was your first computer game? Many recall the classic Minesweeper, a simple yet strategic puzzle.

Playing extensively leads to a set of “strategies”: by recognizing patterns and marking bomb locations, players improve their skills.

A group of former Minesweeper enthusiasts now work at Tencent as “AI Minesweeper experts.” Today, Tencent has released the industry’s first AI Security Attack Matrix , which functions like a Minesweeper guide for AI.

This highly practical AI security framework comprehensively surveys cutting‑edge AI safety research from both academia and industry, listing attack processes and techniques from the attacker’s perspective across all stages of AI development and deployment.

The matrix draws on the mature ATT&CK framework used in network defense, categorizing attacks by maturity levels—“well‑established,” “research‑in‑progress,” and “potential threat.” It expands the analysis beyond algorithms to include design, software dependencies, hardware, supply‑chain, side‑channel, physical attacks, and model architecture security.

Creating this guide was challenging because AI systems combine software and hardware in ways that differ from traditional internet products, leaving engineers with few precedents. The team collected thousands of papers and real‑world security incidents, iterating through multiple versions to align the framework with the AI development lifecycle.

After six months of refinement, the Tencent AI Security Attack Matrix was published, offering a global view of AI‑related security risks and practical defensive recommendations. While it marks a significant step for Tencent’s engineers, the landscape will continue to evolve as new AI attack techniques emerge.