Red teaming
Red teaming is deliberately attacking an AI system before release to find its weaknesses: harmful outputs, jailbreaks, bias, privacy leaks or dangerous capabilities. Testers act like adversaries so real users don't discover the problems first.
In one line, for a 12-year-old
Red teaming is people trying hard to break an AI on purpose, so it can be fixed before everyone uses it.
An example
A bank hires testers to try to make its new customer chatbot reveal other clients' details.
Why it matters to people
Testing is only as good as the testers' range. Including people of different languages, cultures and abilities finds problems a narrow team would miss.