And sampling bias.

Plus they pick and choose numbers for a more drastic headline. “Sensitive” data is a very broad category, I don’t know what criteria they used but that could be as little as someone’s name being mentioned with a “todo” note. The quarter of a percent mentioned as having a “critical” issue I venture is closer to what most people think of when they read the title. Infosec consultants have a bad habit of inflating numbers until actual risks are lost in the noise.