AetherGuardAI/aetherguard-research — reverse-engineered prompt

Build me a Python based security research repo focused on AI infrastructure, with a clean top level README and one finished project called VectorBleed.

The main thing I want is a reproducible proof of concept that shows how a shared vector database could leak data across tenants through embedding similarity or proximity. Make it easy to clone, set up, and run end to end, with clear instructions, sample data or test inputs, scripts for the experiment, and output that helps someone understand what is happening and why it matters.

Keep the tone research oriented and responsible. I do not want a weaponized exploit. I want something that demonstrates the attack surface, explains the assumptions, shows realistic results, and includes practical defenses or mitigation ideas. Organize it so more research projects could be added later in their own folders, but make VectorBleed feel complete right now. If you need to, look up current docs online and fill in sensible details.