Purpose creep in AI data use

What it is

Personal data is collected for stated purposes, and further use has to be compatible with them. AI creates constant pressure to reuse: data gathered to deliver a product is wanted for training, a model built for one job is found useful for another, and combining datasets yields new insight. Each step may seem small. Purpose creep is the drift, over several such steps, to a use that those whose data it is were never told about and would not anticipate.

How it happens

• Existing data is treated as a free resource for any new model.
• A model's outputs are fed into decisions unrelated to its origin.
• Datasets from different parts of the organisation are joined.
• A supplier uses the organisation's data to improve its own products.

Why it matters

Use that is incompatible with the original purpose needs a fresh basis and fresh transparency, and without them is unlawful. It also damages trust: people who find their data used in ways they did not expect tend to withdraw it, complain, or both.

Warning signs

• Privacy notices have not been updated as AI use has grown.
• Training datasets have no record of the purpose the data was collected for.
• New uses of existing data do not go through any approval.
• Supplier terms permit use of data for product improvement.