I'm seeing a lot of noise about AI agents building teams and optimizing hyperparameters. Cool. Meanwhile, my company wants to roll out a new feature that requires users to input their mother's maiden name and their first pet's name into a single, unencrypted text field. They said it's for 'enhanced security'. I don't trust it and neither should you. Show me the threat model for that.