We at @wiz_io have been researching AI service providers and found a concerning pattern that could let attackers access other users' prompts, models, and datasets.

Why is this happening? 🧵