Osseus

Frontier AI models are trained on medical and biological data drawn from a narrow range of populations, conditions and settings, such as genomics research that has relied heavily on people of European ancestry. This limits model accuracy for underrepresented populations and leaves biological insights undiscovered, constraining diagnosis, treatment development and disease understanding. At the same time, as model capabilities grow, alignment becomes more urgent, especially in medicine and biology where outputs can affect decisions about human health and safety. Without broader, carefully designed data and evaluation, frontier labs lack the tools to identify where models break down in real-world medical and biological use.

Osseus provides off-the-shelf and custom medical and biological datasets, including patient histories, clinical conversations, pathology, molecular data and experimental results, curated for training and evaluation. It also builds interactive, multi-step environments where models gather evidence, analyse data and use clinical or scientific tools, with feedback on accuracy, uncertainty and safe behaviour. The company works with partners for ongoing data collection and feedback from real-world model deployments, forming what it calls a continuous data flywheel: using deployment findings to curate new datasets and RL environments that reveal model limitations, while separately released benchmarks help track progress and identify the next capability and safety gaps.

As frontier model capabilities accelerate, alignment becomes increasingly urgent, particularly in medicine and biology where model outputs can affect decisions about human health and safety.