I study corporate innovation strategy using machine learning. In the first chapter, I investigate the strategic-patenting behavior of target firms in the pre-acquisition phase. Specifically, I develop a theory of false signaling in the context of frontier-technology firms’ mergers and acquisitions, where target firms may file “weak” patent applications as misleading innovation signals. Using natural language processing of four million U.S. patent applications, I find that patent applications filed by frontier-technology firms before being acquired are less likely to be granted. However, the negative influence of prospective acquisitions on targets’ patent grants is weakened by technological relatedness and anticipated due-diligence efforts. This chapter contributes to signaling theory, innovation literature, and technological acquisitions literature. In the second chapter, I examine how firms may strategically draft their patent applications by using vague expressions. I find that there is a positive relationship between trade secrets and patent vagueness, but this relationship is weakened by technologically overlapping patent portfolios. To test my predictions, I draw data from pre-grant patent applications and annual reports and employ supervised machine learning. This chapter sheds new light on how firms may make trade-offs between generative and primary appropriability.