European regulators have finalized long-awaited updates to how software-based diagnostic tools are classified, specifically addressing a gap that had left adaptive, continuously-learning algorithms in a gray zone since the original framework was written for static software.
Under the new rules, diagnostic algorithms that update their internal parameters based on incoming data must now declare a “predetermined change control plan” at the time of initial certification, specifying the bounds within which the model is allowed to adapt without triggering a full re-review.
Manufacturers had pushed for this clarity for several years, arguing that the prior framework effectively forced them to choose between freezing model weights after approval or resubmitting for full re-certification after every retraining cycle.
Industry groups have broadly welcomed the update, though several have noted that the documentation burden for the change control plan itself is substantial, and smaller developers may need to budget significant regulatory affairs resources to meet it.
The new classification rules take effect for new submissions within the next reporting cycle, with a transition period for products already on the market.