Disclosure to candidates — telling them AI was used

The direction of travel

Across the jurisdictions Nepal looks to — Europe, the United States, India — the regulatory direction on AI in hiring is converging. The EU AI Act, in force from 2024 with phased application running into 2026 and beyond, classifies AI systems used in recruitment and employee management as “high-risk” and imposes transparency, documentation, and human-oversight obligations on the firms that deploy them. State-level rules in the United States — New York City’s Local Law 144, Illinois’s AI Video Interview Act, California’s amendments to its civil-rights regulations — have already required disclosure to candidates when automated decision tools are used in hiring, with bias audits and notice provisions stacked on top. India’s draft Digital India Act and the DPDP Act 2023 are pushing in the same direction, with growing pressure on employers and platforms to disclose automated processing of personal data.

Nepal has not formalised any of this yet. The 2075 Privacy Act and the draft Personal Data Protection Bill do not specifically address AI in hiring, and the Labour Act 2074 predates the current generation of tools entirely. But the direction is plain. Within a few years, some version of “tell candidates when AI is used in the hiring decision” will be a regulatory expectation in Nepal, whether by primary legislation, by NRB or SEBON directive for regulated sectors, or by court interpretation of the existing privacy framework. The firms that have been disclosing all along will treat that day as a non-event. The firms that have been concealing will spend it rewriting their hiring pages and explaining to their first regulator inquiry why they were not already doing this.

The position that ages well

There is a position on candidate disclosure that holds up under every plausible future and that requires neither dishonesty nor paranoia. It has three parts.

One — tell candidates that AI was used in your hiring process. Not in legalese, not buried in a privacy notice, but in plain language on the careers page, in the first acknowledgement email, or in the interview-confirmation note. “We use AI to help us read CVs and structure our interview notes” is enough at the level of general disclosure. It tells the candidate the tool was in the room.

Two — explain where, specifically. AI was used to draft the JD; AI was used to help rank CVs against a written rubric; AI was used to summarise interview notes for the panel. These are different uses with different stakes, and a candidate reading the explanation should be able to picture the workflow without the firm giving away anything that would compromise it. The level of detail is the same a thoughtful candidate could reconstruct anyway from the patterns in their own application experience.

Three — explain where human judgement closed the loop. This is the part that earns trust. “A member of our HR team reviewed every shortlist against our written hiring rubric before invitations went out. The final hiring decision was made by the line manager and HR director in conversation, with the rubric in front of them.” This sentence does two things at once. It tells the candidate the model did not get to decide. It tells the regulator, when the regulator arrives, that the human-oversight obligation was being met all along.

A practical disclosure template

You do not need a corporate-counsel-drafted document. You need a paragraph that the HR team is comfortable sending and that the candidate can read in under a minute.

A template that works for most Nepali SMEs and mid-sized firms:

“We use AI to help structure our recruitment process. Specifically, it helps us draft job descriptions, organise the notes from interview panels, and produce an initial reading of candidate CVs against our written hiring rubric. A member of our HR team reviews every shortlist before invitations are sent, and the final hiring decision is made by the hiring manager and the HR lead together, with the same rubric in front of them. We do not use AI to make rejection decisions automatically, and no candidate is rejected without a human reviewer having read the CV. If you would like more information about how we use AI in our hiring process, please write to [hr@firm.com.np] and we will share more.”

The template is six sentences. It tells the truth, it pre-empts the most common candidate concerns, and it points the unusually curious candidate to a fuller conversation if they want one. It sits on the careers page and is referenced in the first acknowledgement email. The HR lead reads it once a quarter and updates it when the workflow changes — and the workflow will change.

Candor over evasion

The instinct of many HR leads, faced with the lawyer’s question, is to draft a careful answer. “Our process incorporates a range of evaluation tools, including modern technology, all under the supervision of qualified HR professionals.” This sentence is technically true, says almost nothing, and looks — in the eventual newspaper article — exactly like what it is: an evasion.

Candor reads better in every room. A direct answer to the lawyer’s question — yes, AI was used to help rank CVs against our written rubric; the rubric is attached; a human reviewer revised the shortlist; here is the workpaper for that hiring round — closes the conversation in most cases. The lawyer takes the file to the client, the client reads the rubric, the rubric explains why the rejection happened, and the matter ends. The evasive answer, by contrast, invites a second letter, then a third, then a complaint to whichever authority will take it, then a story in a Nepali tech publication about a startup that would not say whether it used AI on a senior engineer’s CV. None of that needed to happen.

The general rule: vague denials and clever non-answers about AI use will surface. They always surface — through an honest junior, through a former employee, through the discovery process in a labour dispute, through a careers-page screenshot a candidate took before the words quietly changed. When they surface, they look worse than the disclosure would have looked. The cost of the careful answer is paid in the future, and with interest.

The candidate-facing FAQ

For firms that hire at any scale, the disclosure paragraph is a start, not an end. The artefact that closes the loop properly is a one-page candidate-facing FAQ on AI use in hiring, linked from the careers page and shared on request. Five or six questions and answers, written in plain Nepali and English.

What AI tools are used in your hiring process? — names the tools by category, not by vendor secret. What is AI used for, specifically? — JD drafting, CV ranking against rubric, interview-note summarisation, scheduling. What is AI not used for? — making rejection decisions, conducting interviews on its own, scoring candidates’ video or voice without human review. Who reviews the AI’s output? — named role, not named individual. Can I ask not to have AI used on my application? — the answer might be yes (the firm processes your CV manually if you ask), or it might be a regretful no (the firm cannot reliably exclude one application from the workflow at this scale), but the answer exists in writing rather than being improvised each time. How can I get more information about a specific decision? — the contact route, the timeline.

The FAQ is a thirty-minute drafting exercise. It does not appear in marketing materials. It sits in the careers folder, gets sent to anyone who asks, and lives where a regulator, a candidate’s lawyer, or a journalist can find it without having to ask twice. The firms that have this in 2026 will look like the careful ones in 2028. The firms that do not will be writing it under deadline pressure, in the week after their first inquiry.

Check your understanding

Quick check

—

True or false: candidates do not need to be told that AI was used in evaluating their application as long as a human made the final hiring decision.

1. ATrue — the human decision is what matters legally and ethically; the use of AI as an internal aid does not require disclosure.
2. BFalse — concealment of AI use always surfaces eventually, looks worse when it does than disclosure would have, runs against the clear regulatory direction in every jurisdiction Nepal benchmarks against, and erodes the candidate trust that honest disclosure builds.
3. CTrue — disclosing AI use invites lawsuits.
4. DFalse — but only for senior hires, not entry-level ones.

Check answer→

What comes next

This chapter began with the internal containment habits that keep employee data safe and continued with the outward-facing candor that keeps candidate trust intact. The final section pulls back and asks the larger question this entire course has been circling: as AI absorbs more of the mechanical work of HR, what is the shape of the profession that remains? Where does the HR professional’s value sit when the JD writes itself, the CVs rank themselves, and the interview notes summarise themselves? The next section is a short closing essay on exactly that question.