All eukaryotic cells either have mitochondria (animals and fungi) or chloroplasts (plants and most algae), or are derived from ancestors which subsequently succombed to atavism and dispensed with such. It’s important to note that the mitochondrial genome is not technically considered part of its “host” genome and is not involved directly in meiosis which allows the wide variation in sexual reproduction. Variation in the mitochondrial genome from generation to generation comes primarily from mutation, though more recent research has demonstrated that both the action of retroviruses via transduction upon mitochondrial nucleotides, and exchange between nuclear DNA and the mitochondrial genome has occured to the point that in animals nuclear DNA is responsible for many of the protein coding opertions that support and maintain mitochondria. Presumably they’ve exported these functions to their hosts in order to be more efficient at what they do (primarily reducing or synthisizing, and oxydizing adenosine triphosphate) and have a streamlined genome, while the host genome, already full of extraneous and seemingly nonfunctional coding sequences is scarcely affected by taking on the additional functions.
There is also extra-nuclear DNA in free-floating plasmids that are found in some cells. These are generally considered non-coding and not part of the mammalian organism, but this is a role that is being reconsidered by researchers as they discover the more complex than previously suspected behavior and interplay of nucleic acids in cell operations.
Maybe yes, maybe no. The animal nuclear genome has areas of protein coat–the chromatin structures–which regulate access to the nucleic acid and permit only valid “keyholders” to initiate the transcription process. There are many areas of the genome that are not well protected by chromatins, but most are non-critical and apparently non-coding areas that may be evolved specifically as garbage collectors which prevent transcription and reproduction errors. In general, genetic reproduction is a very high fidelity, very reliable process in which errors are measured in terms of fractions of a billion, and most replication errors have no ultimate detrimental result; either the error causes a breakdown of the genome and cell death as the DNA strand unravels and becomes non-functional, or the error is harmlessly absorbed and has no impact on the functioning of the DNA.
Rather than viewing it specifically as a parasitic or competitive process it may just be the bacteria genome randomly searching for better code; looking for complementary phrase that will make the poem of its genome ring that much better. The vast majority of mutations benefit no parties, and make for nasty surprises; most bacteria would like to find a nice niche where they can sap away a little bit of the host’s resources without being detected or attacked. Detectable invections invite the host’s immune system to develop systemic responses which not only destroy existing bacteria but prevent future intrusion by similar organisms, which is bad for everyone.