AI Transparency Notice
Last updated: September 20, 2026
This notice expands Section 4 of our Privacy Policy. It is written for candidates: it says where Arihance Systems Private Limited uses AI, what each model is given, what it is allowed to decide, and how to contest an outcome.
The summary is short. AI assists; a person decides. No model receives your identity, and no code path can decline you.
1. The three rules everything here follows
These are architectural constraints in the product, not policy intentions. Each is enforced by tests that fail the build if the rule is broken.
- No model ever receives personally identifying information. Anonymisation happens at a boundary every AI call passes through, before a request leaves our systems. It is not a setting, and it does not vary by provider, plan or region.
- No automated rejection exists. There is no code path that declines a candidate. An AI output can rank, score, summarise or flag; a named human makes every decision that affects a person, and the record stores which human.
- Every person-affecting decision is logged and explainable, tied to the deciding human, and available to you on request.
2. Where AI is used, and what it gets
| Where | What the model is given | What it may decide |
|---|---|---|
| Scoring your assessment answers | Your answer text and the rubric — no name, no employer, no contact details | Nothing. It produces a score and a rationale that a human can review and a human can override |
| Writing the narrative on your report card | Your anonymised scores | Nothing. It writes the explanation of a result that has already been computed |
| Assessment integrity signals | A single answer, anonymised | Nothing. It produces an advisory likelihood that text was machine-generated. A flag only ever routes to a human |
| Adaptive follow-up questions | Your own previous answer | What to ask next. It cannot score you |
| Generating assessments and job descriptions | The role and skills, written by the employer | Nothing about any candidate |
| Your private practice tools (mock interview, skills-gap review) | The skills you entered | Nothing. These are yours, are never shared with an employer, and never affect your verified score |
Résumé parsing appears in the platform for employers importing their own applicants. It runs on the document they uploaded, and it does not touch the anonymous verified profile that employers evaluate.
3. Which models we use
AI providers are configuration, not code: a task can be pointed at a different model without changing the product. The providers reachable today are named in our Sub-processors list, and that list is the authoritative one — it changes when the configuration does.
What does not change with the provider is Rule 1. Anonymisation happens before the gateway chooses a model, so every provider on that list receives the same PII-free content.
4. Your rights over an AI-assisted outcome
Under the GDPR and UK GDPR (Article 22), India's DPDP Act, the UAE PDPL, the EU AI Act and NYC Local Law 144, you have rights around automated processing. In practice, here is what you can do:
- Ask what happened. Request meaningful information about the logic involved in an outcome and the factors that mattered.
- Contest it. Appeal a score or an integrity flag under our Appeals Policy. A human reviews it within 48 hours, and the outcome is recorded.
- Ask for a human-only route. Where an automated tool would be used to evaluate you, you may ask for an alternative and are told before the tool is used.
- Ask for an accommodation so the assessment measures your ability rather than your circumstances — see the Accommodations Policy.
We do not treat you differently for exercising any of these.
5. What we do not do
- We do not use AI to decline candidates, and no such code path exists to be turned on.
- We do not send your identity, contact details or demographic information to any model.
- We do not train third-party models on your data.
- We do not use AI to infer protected characteristics.
6. Questions
Write to privacy@calhire.com. If your question is about a specific outcome, an appeal gets it in front of a human reviewer with the evidence attached, which a general enquiry cannot.
CalHire publishes its bias-audit position separately at /bias-audit.