Sensitive Images Need a Different Stack

Some image workflows should not be treated like creative playgrounds with nicer fonts.

Medical images can include patients, scans, wounds, forms, identifiers, and clinical context. Legal images can include evidence, documents, people, locations, case strategy, and privileged material. Education images can include children, classrooms, accessibility needs, safeguarding concerns, and family context.

These are not ordinary uploads. They come with duties. Sometimes legal duties. Sometimes contractual duties. Sometimes moral duties that are not less real just because nobody put them in a dropdown.

AI can still be useful in these domains. It can redact, enhance, generate training material, create diagrams, anonymise examples, simulate layouts, and improve internal communication. But the stack needs to be chosen for the risk, not the demo.

The wrong stack is a general-purpose image tool with unclear retention, broad service-improvement language, support access to raw files, public sharing features, and no clean way to prove deletion. That may be fine for making a fantasy wallpaper. It is not fine for a patient image, a classroom photo, or a legal exhibit.

A better stack minimises content exposure. It should support private processing, no training on customer content, short-lived source files, restricted access, clear audit boundaries, and ideally the ability to keep sensitive work inside infrastructure the organisation controls. It should also make redaction and de-identification easier, not an afterthought handled by a tired human at 6pm.

There is an honesty requirement too. AI output in these domains can be wrong in ways that matter. Enhancing a medical image is not the same as making a product shot prettier. Generating legal visuals is not the same as proving facts. Education images involving children require caution even when the task looks harmless.

So the product copy should be restrained. Do not promise miracles. Promise controls. Explain retention. Explain provider access. Explain whether human review happens. Explain what the tool is and is not suitable for.

Medical, legal, and education teams do not need “move fast and upload things.” They need tools that respect the context carried by images. Same AI category, different stack.

Further reading: The biometric data hiding in ordinary product photos, The privacy policy questions your AI image feature must survive, and Unexposed data storage.