The challenges raised by the introduction of autonomy in mission-driven and safety-critical aerospace and defense applications is driving a strong push towards the utilization of AI/ML-based techniques. Example of such applications are formation flying and teaming, man-unmanned teaming, collision avoidance, aerial infrastructure inspection, last mile delivery, etc. Standardization bodies, such as SAE and EUROCAE, with the “Artificial Intelligence in Aeronautical Systems: Statement of Concerns”, have clearly identified that it is now necessary to incorporate existing work to produce a standard that provides the necessary accommodation to support the integration of Machine Learning (ML) enabled sub-systems into safety-critical aeronautics software, hardware, and system development. Addressing these challenges, this paper presents a simulation-based framework for the development and validation of mission-critical applications including AI/ML-based components within a safety-critical function implemented in a model-based environment. The framework is based on a mission-simulation backbone that integrates the software model being developed together with aircraft and physics-based sensor simulation (camera, lidar and radar) and connects to AI training environments through a standard OpenAI Gym interface. As an initial step in the system development workflow, the Operational Design Domain (OOD) is defined, and scenarios are created to train and validate the application that typically includes both traditionally developed components and AI/ML-based components. The AI/ML training process is based on simulating these scenarios within the framework while varying their parameters according to their probability distribution. This process accommodates supervised learning, typically used for perception components, as well as reinforcement learning, typically used for decision making or flight planning components. Sensitivity analysis is used to characterize the resulting neural network. Once trained and validated, the AI/ML component is integrated within the overall application design model and the simulation framework can again be used to perform reliability analysis and calculate a probability of failure, thus confirming the performance and safety objectives of the application over its ODD or triggering further training or redesign activities that could be necessary to meet the objectives. Finally, the embedded code is generated from the software model by using a certified code generator. Overall, the simulation framework enables AI/ML-based decision making for autonomous systems for applications in complex and uncertain environments, supports autonomous and pilot-assistance systems while guaranteeing system safety, and supports the design of efficient systems performance such as energy-aware trajectories or area-coverage maximization. The approach is illustrated by two use cases, including formation flying with a piloted lead aircraft followed by autonomous vehicles, and a flight maneuver where a UAV must avoid collision with an intruder aircraft. Users of the simulation framework include both the system developers and the system operators that can build and use digital and executable reference models that cover the mission and the vehicle behavior. This allows to include operational experience into a digital validation environment that system developers can leverage to assess their design and implementation. Vice versa, simulating the system in an actual mission environment allows system operators to better understand the system behavior and provide earlier and more accurate feedback to the system developers.