MIT researchers demonstrated that branch-predictor defences can be re-poisoned in the brief interval between being cleared and being used. This is a local attack, not a drive-by remote compromise. At Black Hat USA on 6 August, MIT researchers Daniël Trujillo…