---
title: "Comment by Kartik N V J K on \"Your AI Assistant Downloads Code From Strangers. Here's Why That's a Problem\" | Hashnode"
description: "The \"harmless parts, dangerous combination\" point (read-files tool plus send-email tool equals exfiltration path) is the risk I think teams underrate the most,"
url: "https://hashnode.com/posts/why-ai-assistants-download-code-from-strangers/6a5da6803ef9671a60d23160/comment/6a5e4c488942a4f8dbfaa726"
---

[Discussion](https://hashnode.com/posts/why-ai-assistants-download-code-from-strangers/6a5da6803ef9671a60d23160)

Comment [Your AI Assistant Downloads Code From Strangers. Here's Why That's a Problem](https://hashnode.com/posts/why-ai-assistants-download-code-from-strangers/6a5da6803ef9671a60d23160)

[Kartik N V J K](https://hashnode.com/@kartik-nvj)

[Jul 20](https://hashnode.com/posts/why-ai-assistants-download-code-from-strangers/6a5da6803ef9671a60d23160/comment/6a5e4c488942a4f8dbfaa726)

The "harmless parts, dangerous combination" point (read-files tool plus send-email tool equals exfiltration path) is the risk I think teams underrate the most, because each capability passes review in isolation. The rename-trick detail is a good reminder that inspecting by file extension is theater; content inspection of the serialized model is the only thing that holds. I would be curious whether you scope tool combinations at the policy layer too, since even a fully inspected model can be talked into chaining two safe tools by an injected instruction in retrieved content.

[Mohit Kumar](https://hashnode.com/@mkd)

[Jul 20](https://hashnode.com/posts/why-ai-assistants-download-code-from-strangers/6a5da6803ef9671a60d23160/comment/6a5e4c488942a4f8dbfaa726#reply-6a5e52a8f74b4ed5359b8627)

Thanks, Kartik! That's a great observation. You're absolutely right that inspecting individual components isn't sufficient if the agent can still be induced to chain otherwise safe tools into a risky workflow. That's actually one of the motivations behind Warden in Bulwark. It analyzes agent assemblies, capability combinations, and privilege boundaries rather than evaluating tools in isolation. The policy layer and "safe parts, dangerous combinations" problem is exactly where I think AI security needs to evolve. There's definitely more work to do around prompt-injection-driven tool orchestration, and it's an area I want to keep expanding. Thanks for raising it!
