---
title: "Discussion on \"AI Threat Modelling: Understanding Attack Surfaces in Modern AI Systems\" | Hashnode"
description: "Discussion on \"AI Threat Modelling: Understanding Attack Surfaces in Modern AI Systems\". AI systems don’t fail like traditional software. Learn how attackers exploit inputs, data, and outputs—and how to think like a security engineer."
url: "https://hashnode.com/posts/ai-threat-modelling-part-1/69c71ef37cf270651071a60c"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [AI Threat Modelling: Understanding Attack Surfaces in Modern AI Systems](https://cyberfreak.hashnode.dev/ai-threat-modelling-part-1?utm_source=hashnode&utm_medium=feed)

AI systems don’t fail like traditional software. Learn how attackers exploit inputs, data, and outputs—and how to think like a security engineer.

Mar 28 · 4 min read · cyberfreak.hashnode.dev  
Read the article

-   [#ai-security](https://hashnode.com/tag/ai-security)
-   [#ai-threat-modeling](https://hashnode.com/tag/ai-threat-modeling)
-   [#machine-learning-security](https://hashnode.com/tag/machine-learning-security)
-   [#llm-security](https://hashnode.com/tag/llm-security)
-   [#application-security](https://hashnode.com/tag/application-security)
-   [#cybersecurity-fundamentals](https://hashnode.com/tag/cybersecurity-fundamentals)
-   [#ai-attack-surfaces](https://hashnode.com/tag/ai-attack-surfaces)
-   [#data-security](https://hashnode.com/tag/data-security)
-   [#model-security](https://hashnode.com/tag/model-security)
-   [#prompt-injection](https://hashnode.com/tag/prompt-injection)

## Written by

[CyberFreak999](https://hashnode.com/@CyberFreak999)

@CyberFreak999 · Rivers State, Nigeria

  
[View profile](https://hashnode.com/@CyberFreak999)

Backend Developer | Transitioning to Cybersecurity

## Responses 1

Markdown supported · ⌘↵ to post

[Aamer Mehaisi](https://hashnode.com/@mehaisi)

[Mar 28](https://hashnode.com/posts/ai-threat-modelling-part-1/69c71ef37cf270651071a60c/comment/69c71f2819efefe03c09e7f1)

The chatbot leaking internal emails example captures the core tension: AI systems have wider attack surfaces than traditional software because their behavior is shaped by training data, not just code.

Three threat categories worth highlighting:

1.  Indirect prompt injection via retrieval - RAG pipelines are attack vectors if untrusted documents enter the knowledge base. Models do not distinguish between user instruction and retrieved context at inference time.
    
2.  Model weight tampering - For local deployments, model files ARE code. Supply chain attacks that modify weights create backdoors that survive code review.
    
3.  Training data extraction - Attackers do not need training data access if they can craft inputs that cause the model to regurgitate training examples.
    

The article focuses on inference-time threats, but the attack surface spans training, fine-tuning, and deployment. Threat modeling for AI needs the full pipeline, not just the API boundary.
