Research in the area of security for Wireless Sensor Networks over the past two decades has yielded many interesting findings. We focus on the topic of (re-)securing link keys between sensor nodes through so-called secrecy amplification (SA) protocols. Crowdsourcing is at the very heart of these SA protocols. Not only do SA protocols work wonders even for low-level constrained nodes with no tamper resistance, they exhibit astonishing performance in networks under significant attacker control. Our work shows that even when 50% of all network links are compromised, SA protocols can re-secure over 90% of the link keys through an intriguingly simple crowdsourcing mechanism. These protocols allow us to re-take control without any broadly coordinated cooperation, without knowledge of the compromised links, with only very limited knowledge of each particular network node and independently of decisions made by other nodes. Our article first outlines the principles of and presents existing approaches to SA, introducing most of the important related concepts, then presents novel conclusive results for a realistic attacker model parametrised by attacker behaviour and capabilities. We undertook this work using two very different simulators, and we present here the results of analyses and detailed comparisons that have not previously been available. Finally, we report the first real, non-simulated network test results for the most attractive SA protocol, our implementations of which are available as open-source code for two platforms: Arduino and TinyOS. This work demonstrates the practical usability (and the attractive performance) of SA, serving as a ripe technology enabler for (among others) networks with many potentially compromised low-level devices.