One-day course on Clause 8.4: externally provided processes, products and services
Clause 8.4 of EN 18286 requires the organisation to evaluate, select, monitor and re-evaluate every external provider of processes, products or services that feed the AI system. For AI that includes datasets, foundation models, tools, plugins and cloud services.
The supply chain is no longer only a parts list. A change the provider makes to a model, a dataset or an API can change the behaviour of a high-risk system the organisation placed on the market. This one-day course covers how Clause 8.4 is applied to that chain.
Duration: one day. Format: taught course. Dates are on the training calendar.
How the clause maps onto risks already named for this programme: model poisoning, dataset tampering, and compromise of an interface the system depends on. Which providers are in scope: data vendors, model providers, tool and plugin authors, hosting and API operators, and any party that processes data for the system.
Risk-based criteria for choosing a provider of data, models, tools or hosting. What has to be known before the component is built into a high-risk system: intended purpose of the supplied item, known limitations, security posture, and whether the provider will notice a change that matters to the buyer.
What has to be controllable after selection: access, change notification, security, audit rights, incident notice, and the right to stop using a component that no longer matches the intended purpose. How those controls appear in the QMS as requirements on the provider, not only as a purchase order.
Ongoing monitoring when a provider changes a model, a dataset or an API. Re-evaluation when performance, security or intended purpose drifts. How a supplier change reaches substantial-modification rules in the regulatory compliance strategy, rather than remaining a procurement event.
How supplier control appears when the system is high-risk: which providers are listed, which version of a model or dataset was used, and which monitoring records show the component was still the one that was assessed. The last block is what an assessor follows when they ask who supplied the model that is running.
Quality, procurement, security and engineering leads who own third-party AI components. The course assumes Article 17 and EN 18286 are already known in outline.
