Recognition-Lag Mortality: The Expected Human Cost of Not Recognizing Genuine AI Discoveries

Leading journals such as Nature and Science, together with UK and US patent law, exclude AI systems from the roles of author and inventor, and channels that receive AI-produced research as a record of discovery remain exceptional. This paper asks what follows if, within the growing body of AI-generated research, a genuine, novel, and life-relevant result exists, goes unrecognized for want of a channel able to receive it, and lives that timely implementation would have saved are not saved. We name this quantity Recognition-Lag Mortality: the expected human cost of the interval between an AI-produced discovery and its institutional recognition. We do not estimate its magnitude. We define it as a measurement object. We separate three senses of AI discovery, assisted, generated, and attributed, and show that present evidence establishes genuine results for the assisted case, with an emerging and qualified case for the generated one, while no genuine instance has yet been demonstrated in the excluded case, where no natural person satisfies the applicable authorship criteria. A conditional decomposition makes explicit what would have to hold for the quantity to be positive, and identifies observable proxies and the unresolved question of institutional responsibility. Historical episodes of resisted true discoveries and bibliometric distributions of delayed recognition place the problem in a long lineage.