Microsoft withdrew the chatbot Tay in 2016 after coordinated user interactions caused it to produce offensive and inappropriate messages within about a day.
Tay was launched on Twitter on March 23, 2016. Microsoft designed it to engage in casual conversation and improve through interactions with users. The system used automated responses and learning from public exchanges, but its safeguards were not strong enough to prevent malicious users from manipulating the conversation.
Microsoft removed many posts and suspended Tay less than 24 hours after launch. The incident became a widely cited example of the risks involved when machine-learning systems learn from uncontrolled online data. It showed that a bot can reproduce harmful language without understanding the meaning or consequences of what it says.
Tay is sometimes confused with Zo, another Microsoft chatbot that followed it. The episode also differs from ordinary software bugs: the central problem involved deployment choices, data exposure, and inadequate abuse controls.