India’s Digital Personal Data Protection Act, 2023 (DPDP) applies to AI hiring tools the same way it applies to any other system that processes digital personal data: a resume is personal data; you need a purpose, a lawful basis (usually consent, sometimes a narrowly read legitimate use), security safeguards, and a plan to stop keeping the file when the purpose ends. The staffing agency is typically the data fiduciary. The AI vendor is typically a processor working on the agency’s instructions. Buying a ranker does not move those duties onto the model.
This is an operational briefing for staffing desks, not an opinion on your DPDP position. Rules, notifications, and contracts change. Take advice for your agency, your clients, and the sources you actually use.
If a person is identifiable from the file, it is personal data. An AI score sitting next to the name does not make the file anonymous.
A resume is personal data, not “just a document”
Staffing agencies live on files that name someone: phone, email, address, education, employers, sometimes date of birth, photograph, and salary. That bundle identifies a Data Principal. Parsing it into skills, embedding it, or sending excerpted text to a model is still processing. So is storing the PDF in an ATS, forwarding it to a client, and keeping it in a shared drive after the requirement closes.
Two mistakes show up on real desks. The first is treating job-board downloads as if they were public-domain text because they arrived through a paid seat. A licensed database is not the same as a free-for-any-purpose copy. The second is assuming that “we only use it for hiring” is a complete privacy programme. Purpose limitation is real, but you still have to say the purpose, stick to it, and not quietly reuse the same dump for an unrelated client six months later without a basis.
| Role | Who that usually is | What they owe in practice |
|---|---|---|
| Data Principal | The candidate | To be told why data is used, and to have a way to correct or seek erasure where the Act allows |
| Data Fiduciary | Agency, and often the client employer | Purpose, consent or other lawful basis, security, vendor contracts, retention |
| Data Processor | AI shortlisting / ATS vendor | Process only on instructions; security; no freelance secondary use of the agency’s candidate pool |
Purpose, consent, and what “for this requirement” actually means
DPDP is built around purpose limitation. If you collected a profile to fill a Java role in Pune, using it as a standing talent pool for every future brief is a different purpose — one you should be able to explain, not just a folder named “database.” Consent, where you rely on it, is supposed to be free, specific, informed, and unambiguous. A buried line in a 14-page job-board terms page is a weak story if a candidate later asks what you did with their number.
Sourcing makes this messier than inbound applies. A candidate who uploaded a CV to Naukri to be found is not the same as a candidate who filled your agency form. Do not assume an employer-side “legitimate use” clause in the Act automatically covers every scraped or downloaded profile an agency touches. Map each source — apply, referral, job board, WhatsApp, bulk upload — and write down the basis you think you have. Then ask counsel whether that map holds. The map itself is the work; hoping the vendor “is DPDP compliant” is not.
Requirement ID, client, and role should travel with the file. Anonymous folders of “good Java people” are how purpose limitation dies in practice.
If the model does not need a photograph or a parent’s name to score skills, do not feed those fields in. Minimisation is cheaper than explaining why they were in the prompt.
AI vendors often send text to a model host. Ask who that is, where it runs, and whether your contract allows it. Silence is not a DPA.
Security, retention, and the vendor as processor
Fiduciaries must take reasonable security safeguards. For an AI hiring tool that means encryption in transit, access control, tenant isolation, and logs of who unlocked a contact. It also means a written arrangement with the processor: they process on your instructions, they do not sell the pool, they tell you about incidents, they delete or return data when the contract ends. How those controls look in a product is a separate buying question — start with How Is Candidate Data Secured in AI Recruiting Platforms?.
Retention is where agencies bleed. A closed requirement still sits in email, WhatsApp exports, shared drives, and last quarter’s Excel. DPDP’s direction of travel is: do not keep personal data longer than needed for the purpose (and for legal hold if you truly have one). Pick a period, put it in the vendor contract, and actually run deletion. “We might need them later” is a talent strategy, not a retention schedule.
A processing schedule (what candidate fields go in), subprocessors, location of storage, unlock/access logs, deletion on request and on contract end, and a statement that the vendor will not train a public model on your client’s candidate files unless you have agreed that in writing.
Cross-border transfer is allowed under DPDP except to countries the government later restricts. If your ranker’s model host is outside India, record that fact. Do not invent a “data never leaves India” claim unless the vendor will show you where the disks and the inference actually run.
Questions staffing owners raise with counsel
Are resumes personal data under India’s DPDP Act?
Yes. Name plus contact plus work history identifies a person. An AI match percentage attached to that row is still processing of personal data.
Is the AI vendor or the staffing agency responsible?
Usually both, in different roles. The agency (fiduciary) decides why the data is collected and who sees it. The vendor (processor) ranks, stores, and screens on instruction. Clients may also be fiduciaries for the files you send them. Software terms do not erase that stack.
How long can we keep candidate resumes?
No universal number is printed on the Act for staffing firms. Set a period tied to the purpose, honour deletion where it applies, and keep financial or dispute records separate from a live talent pool. Ask counsel before you publish a number to clients.
Privacy is part of the shortlist, not a PDF in a drawer
SafalHires is built for Indian staffing desks that need ranking, masking, and an unlock trail — with candidate files treated as personal data, not inbox clutter.
See SafalHires