The same phenomenon appears to occur in the singular form. I very frequently hear Übung(K) realised with a voiceless stop (e.g. [k]) instead of the expected velar nasal [ŋ], even in relatively formal registers (e.g. on public broadcasters such as ZDF or ARD, including by highly educated speakers and media presenters like Wagenknecht, Maischberger). In such cases, the canonical [ŋ] seems to be replaced by a voiceless obstruent, reminiscent of the process of Auslautverhärtung (final devoicing). Some of the exceptions being Markus Lanz as a media presenter, but he was born in Südtirol and then he moved and studied in a Sprechakademie in Hamburg how to correctly pronounce in Hochdeustch , another being for example Cornelius Beck in perfekt Hochdeutsch sprechen in his Tübinger Hörakademie . They both pronounce consistently in all contexts "ng [ŋ], Gott sei Dank lol!
It remains unclear whether this should be analysed as a form of hypercorrection or as a misinterpretation/overgeneralisation of final devoicing patterns. This is somewhat unexpected, given that phonological representations are typically stabilised early in acquisition (i.e. during early childhood).
Alternatively, the phenomenon may be driven by perceptual factors (i.e. speakers reproducing what they believe they hear), dialectal influence, or articulatory difficulty. In particular, the production of the velar nasal [ŋ] may be less stable for some speakers, especially in complex phonotactic environments. From a learner perspective, this is also evident: in my case (given my L1 background), [ŋ] is not especially problematic in word-final position, but becomes more difficult when followed by another segment, particularly a vowel.
For example, in Finger, whose standard pronunciation in German is [ˈfɪŋɐ] (without a following [g] or [k]), I tend to produce a form closer to English /ˈfɪŋɡə/ with a hint of G , i.e. with a hint of G , in contrast to words like king /ˈkɪŋ/, where [ŋ] occurs in absolute final position and easier to pronounce for me .