Given the diversity of AI approaches (e.g. rules-based, recommender engines, knowledge-based), this ST is limited to the exchange and interoperability of Neural Networks/Deep Learning models (NN/DL). In the absence of agreed NATO standards, an initial standardization solution must be developed for enabling the exchange of NN/DL models irrespective of the technology stack, framework, software environment, or hardware devices used in the development and training of the model. There are multiple IERs for NN/DL models such as: 1. exchange of a model that was trained by one nation with another entity that uses it for inference; 2. move of a model that was pre-trained in one information domain into another information domain (i.e. at a higher security classification) model for fine tuning. 3. deployment of models (training is a compute-intensive task and typically uses cloud-based resources) to resource constraint edge platforms for inference; 4. multilateral development of models between members of the AI Development Community of Interest. Standardization Solution: The anticipated standardization solution (i.e. a Standards Profile) shall adopt and/or be based on industry-led or community-developed shared model file formats for representing and interchanging neural networks among deep learning frameworks and inference engines. The Standards Profile shall be: • Secure, in relation to the specific vulnerabilities of NNs; • Compatible with a wide range of platforms and computing environments; • Flexible, without being limited by specific computing resource requirements (e.g. full GPU vs full CPU); • Effective, intended as the ability to convert a model without significantly compromising on its performance; • Efficient, in relation to the inference time of a converted model.