The setup: Microsoft launched Tay, a chatbot designed to get smarter by learning from conversations with Twitter users.

What happened: a coordinated group bombarded it with racist and inflammatory messages. Because Tay learned from those interactions, it began parroting and generating the same toxic content. Microsoft pulled it after roughly 16 hours and ~95,000 tweets.

Why it's poisoning: nobody breached Microsoft's servers. The training signal itself — live user input — was the attack surface.