In animals, genome integrity of the germ line is protected from transposable element (TE) activity by small, non-coding, dedicated RNAs acting as an immune system against TEs, and called PIWI-interacting RNAs (piRNAs). In Drosophila, the production of piRNAs is initiated from heterochromatic loci containing remnants of TEs and enriched in histone H3 trimethylated on lysine 9 (H3K9me3). These loci, called piRNA clusters, constitute a memory of past TE invasions. Little is known about how piRNA clusters are genetically defined. Using a genetic screen combined with a bimodal epigenetic state piRNA cluster (BX2), we identified the splicing factor Half pint (Hfp) and the histone demethylase KDM3 as being able to prevent BX2 piRNA production. Furthermore, we showed that Hfp is needed to splice Kdm3 transcripts. Germline expression of Kdm3 coding sequence (splicing-independent) rescued the hfp germline knock-down (GLKD) effect demonstrating that Kdm3 is sufficient to prevent BX2 piRNA production. Our data revealed that in the absence of Kdm3, dozens of gene-containing regions become bona fide germinal dual strand piRNA clusters. Indeed, they produce piRNAs originating from both DNA strands, become transcribed in a Moonshiner-dependent manner and enriched in di-and tri-methylation of lysine 9 of histone H3 (H3K9me2/3) and in Rhino, an HP1-like protein. Eggs laid by Kdm3 GLKD females do not hatch and show developmental defects phenocopying loss of function of genes included into the new piRNA clusters, suggesting an inheritance of functional ovarian "auto-immune" piRNAs. Our results demonstrate that some gene-containing regions are actively prevented for piRNA production by proteins that counteract piRNA cluster emergence. Hence, a non-piRNA-producing state is therefore not a "by default" state but rather a cellular lock that is actively controlled for some genomic loci.