A genome language model is an AI system trained on DNA or RNA sequences so it can learn biological patterns and generate new sequence candidates. Unlike simple motif search, these models try to capture long-range constraints across a sequence, which helps them produce outputs that remain biologically plausible.
In cyber security and biosecurity, this matters because generation is a dual-use capability. The same model that can help design useful research tools or study evolution can also propose novel nucleic-acid sequences that may evade naive pattern-matching filters. That is why defenses now include sequence screening, access controls, human review, logging, and red-teaming. The key risk is not only known hazardous sequences, but also plausible new ones that a model can create at scale.