Cura 1T Uses a 'Human-Gated Self-Evolution Loop' to Train a Healthcare Agent Without Capability Regressions

arXiv / HuggingFace Daily Papers (36 upvotes)·medium signal

actAVA AI's Cura 1T (arXiv 2607.15314, July 15) targets patient consultation, multimodal clinical reasoning, interactive diagnosis and EHR tool use in one model. The training method is the interesting part for builders outside medicine: rather than one generic domain update, agents plan targeted capability improvements, train, evaluate, and refine data from observed failures — explicitly to stop gains in one area from degrading another. The authors report ranking at or near the top among frontier baselines on healthcare evals while staying competitive on general reasoning and agentic benchmarks.