This seminar will present recent results from the ATLAS Collaboration in searches for non-conventional jet signatures motivated by extensions of the Standard Model that include dark QCD-like dynamics, including one result that uses data from LHC Run 3. The analyses target two distinct topologies: semi-visible jets, which consist of a mixture of visible Standard Model hadrons and invisible dark-sector particles, and emerging jets, characterized by displaced vertices resulting from long-lived dark hadrons decaying back into Standard Model particles. To enhance sensitivity to these non-standard signatures, advanced machine learning techniques, including transformers trained on charged-particle track features, are used to discriminate signal-like jets from QCD backgrounds. A model-independent ``anomaly detection'' search strategy was also implemented to cover a wide range of the dark-sector parameter space. These results represent a significant step in probing the structure of hidden sectors and contribute to the broader effort to uncover the particle nature of dark matter.