Anonymous online groups such as imageboards have increasing cultural influence. Recently, they have been connected with far-right political movements. This mixed-methods study investigates politics on Overboard, a Finnish imageboard. We use a convolutional neural network to learn linguistic features of the community’s own understanding of politics, studying two large text corpora, collected in 2014–2015 and 2018–2019. This enables us to find political messages in nominally non-political subforums and discount non-political “noise”—finding the “needles in the haystack.” We quantify the prevalence of political talk on Overboard, assess its themes using topic modeling, and evaluate changes in their popularity. Finally, we qualitatively analyze the style of Overboard. We find that around one-tenth of messages on Overboard are identifiable as “political.” Often, but not univocally, they voice far-right opinions, usually somewhat ironically. The prevalence of far-right themes has increased, likely because of importing global imageboard culture and in parallel with the increased popularity of nationalist-right politics in the broader Finnish public sphere. In terms of group style, the strong boundary between members and outsiders, together with the ironic and cynical speech norms, creates a bond between members. Such a group style lends itself to politicizing the collective.