This OECD paper provides a structured method for classifying AI systems by context, data and inputs, model characteristics, task, output, and potential impacts on people and organizations. It is not a process safety document, but it is highly relevant to PSM because AI risks depend heavily on deployment context. The same algorithm may be low risk in document search but high consequence if used to support operating decisions, safety-critical maintenance prioritization, or hazard evaluation. The OECD classification framework can help PSM practitioners characterize AI use cases before assigning governance controls, human review requirements, validation depth, and documentation expectations. It is a useful front-end screening tool for determining whether an AI application belongs in routine IT governance or in a stricter safety assurance process.
Note that this was published 10 months BEFORE the public release of ChatGPT. So the models and products discussed in this document are not the foundation models commonly used today.
