The DPDP Act and campus AI: what actually changes for colleges

Discussions of AI privacy in education are usually imported wholesale from the United States, where FERPA sets the terms. That framing does not transfer. Indian institutions answer to the Digital Personal Data Protection Act, 2023 and the rules made under it, and the obligations sit in different places.

The first thing to establish is role. A college that collects and processes student personal data is a data fiduciary. That status is not delegable. If the institution routes student data through a third-party assistant, the institution remains accountable for what happens to it — not the vendor, and certainly not the individual faculty member who pasted it in.

Purpose limitation is where most campus AI deployments quietly fail. Personal data collected for admission, examination or placement was collected for those purposes. Feeding the same records into a general assistant, which may retain them, use them to improve a shared model, or process them outside the boundary the institution described, is a different purpose. Consent obtained at admission does not stretch to cover it.

Consent itself is stricter than institutions expect. It must be free, specific, informed and unambiguous, given by clear affirmative action, and it must be as easy to withdraw as it was to give. A clause buried in an admission form does not meet that standard. Where the institution enrols students under eighteen, the requirements tighten further, including verifiable parental consent and constraints on behavioural tracking.

Then there are the rights the institution has to be able to honour. A data principal can ask what is held, ask for correction, and ask for erasure. Erasure is the one that exposes architecture. If institutional data has been absorbed into a shared model, there is no meaningful way to remove it — the obligation and the architecture are simply incompatible.

The practical consequence is that data residency and deletion stop being procurement preferences and become compliance requirements. If the index sits on campus or in a dedicated Indian tenancy, if nothing is used to train shared models, and if documents can genuinely be deleted, then the obligations above are answerable. If not, the institution is carrying a liability it cannot discharge.

None of this argues against AI on campus. It argues for deploying it inside a boundary the institution controls, which is a design decision made once, at the start, rather than a policy written afterwards.