Differential privacy
A mathematical guarantee that the output of an analysis looks almost the same whether or not any one person's data is included.
The definition came in 2006 from Cynthia Dwork, Frank McSherry, Kobbi Nissim and Adam Smith, and won the Gödel Prize in 2017. It answered a recurring embarrassment: "anonymized" datasets kept getting linked back to real people, Netflix's movie ratings in 2008 being the famous case. Instead of promising that the data is de-identified, you promise something about the algorithm: calibrated noise is added to answers, and the parameter ε bounds how much any single record can shift the result.
Each query spends part of a privacy budget, and once it's gone you stop asking. Apple, Google and the US Census Bureau (2020) all use it. Statisticians complained about the noise; the maths was unmoved.