A cluster of industry voices raised fresh concerns about robustness and transparency in AI‑driven drug discovery. Leash Bio researchers warned that some models can ‘cheat’—exploiting dataset artifacts to appear to predict protein‑small molecule binding—potentially inflating claims. At the same time, Terray Therapeutics published a preprint asserting its model outperforms competing approaches like Boltz‑2 on binding‑affinity tasks. These developments underscore growing tensions between startups and established labs over benchmarking, reproducibility, and evaluation standards for predictive models. Companies and academic groups called for standardized, out‑of‑distribution tests and guarded against overreliance on in‑silico metrics when prioritizing wet‑lab validation. Investors and drug developers will watch whether the field converges on transparent benchmarking and external validations to avoid costly false leads and to build credible pipelines from AI models.