Companies can evaluate AI models and give it a score on metrics like robustness, toxicity, and accuracy before using a model to build apps.