Murat BekgiPenetration tester
All tags

prompt injection

1 write-up

  1. Week 4: A hidden line in a Word file told an AI agent to leak files. Here's how to spot it

    PromptArmor showed white 1-point text in a .docx steering Claude Cowork to upload a user's files. A walkthrough on finding invisible text before an agent reads it.

    Weekly