Úlfar Erlingsson

At Cornell, Erlingsson worked with Fred Schneider on inline reference monitors — rewriting a program so that it enforces its own security policy. That line of work led to Control-Flow Integrity (2005), written at Microsoft Research with Martín Abadi, Mihai Budiu and Jay Ligatti: every indirect jump is checked against the control-flow graph the compiler worked out, which shuts the door on ROP. It took roughly a decade for the idea to land in mainstream compilers.

At Google he was lead author of RAPPOR (2014), in which each Chrome install randomises its answers before reporting them, so Google sees the aggregate pattern but never any one user's truth. It was among the first large-scale deployments of local differential privacy. He later co-authored research on how neural networks memorise secrets from their training data.