They fitted some penguins with chemical-sensing silicone passive samplers.
It's almost like legislators saw that the machines could ostensibly detect 4 parts-per-trillion and decided that should be the limit without continuing to read the machine manual to describe the reduction in accuracy at that threshold level.
The levels in this test were close to this threshold and there was one outlier sample that severely changed the average results. The testing methodology also involved several laboratory steps where contamination could have occurred.