Many extensions of the standard model can give rise to tau leptons produced with non-conventional signatures in the detector. Certain long-lived particles may decay into tau leptons displaced significantly from the primary proton-proton interaction vertex. Traditional tau reconstruction and identification methods are not designed for such displaced signatures, necessitating the development of specialized techniques. This note details an implementation of a machine learning approach that employs the ParticleNet architecture to tag these displaced tau leptons effectively. The newly developed tagger demonstrates strong performance in distinguishing tau leptons from quark and gluon jets, particularly in events with large displacements of 10-100 cm, where it has a typical efficiency of 70-80\% at a misidentification probability of $1.8\times10^{-4}$. Its performance is validated with proton-proton data collected in 2018 by the CMS detector.