Active galactic nuclei (AGN) are signposts of the luminous growth of supermassive black holes. During this phase, strong winds can impact the formation of new stars in a galaxy. The short lifetime of the AGN phase makes them very rare objects. Moreover, they are often found with varying methods across the electromagnetic spectrum, with very little overlap between the selected galaxy samples. In tandem with the traditional geometrical paradigm that gives rise to the AGN types (obscured versus unobscured), pieces of evidence point towards an evolutionary approach linking the various AGN populations and also the host galaxy. Upcoming surveys such as Euclid and LSST will offer a significant boost in galaxy and AGN studies due to the unparalleled sample size of billions of detections. Machine-learning techniques are arguably a very efficient approach to selecting AGN, next to traditional AGN selection methods. Such methods can be trained to identify known classes of objects, but they can also be used for novelty and anomaly detection. I will present recent results and future prospects of machine-learning AGN identification methods in anticipation of Euclid.