Context minimisation is the practice of assembling the least context sufficient for a task — restricting how much, how sensitive and how broad the retrieved material is — applying the data-minimisation principle to AI at runtime rather than at collection.
What is context minimisation?
Most retrieval pipelines are tuned for recall, on the assumption that more context produces better answers. It also produces more exposure, higher cost and more opportunity for inference. Context minimisation inverts the default: retrieve narrowly, expand only when the task demonstrably requires it, and treat each expansion as a policy decision. It is the most direct operational expression of GDPR data minimisation inside an AI system, and it reduces token spend as a side effect.