AI Transparency Notice

This notice tells candidates and employer/recruiter users how AI is used on the Kempian Platform and where human decision-making remains required.

> *This document is part of Kempian's Trust Center documentation. It is reviewed periodically and does not constitute legal advice. Draft v0.2 — pending final legal review before publication.*

1. Purpose of This Notice

This notice explains, in plain language, how Adept AI Inc. ("Kempian," "we," "us") uses artificial intelligence (AI) on the Platform. It covers what each AI component does, what data it touches, and — most importantly — where human decision-making remains required. It is written for both candidates and employer/recruiter users of the Platform.

This is a factual disclosure, not a marketing description. It describes Kempian's AI controls and commitments honestly, without overstating what is already guaranteed. Kempian's recruitment and candidate-matching functionality is treated as a high-risk AI use under applicable law (see Section 11). This notice is part of how we meet our transparency obligations to the people affected by that use.

This notice does not replace the Kempian Privacy Policy or Terms and Conditions, which govern the underlying lawful basis for processing and the Platform's contractual terms. It should be read alongside those documents.

2. Where AI Is Used on Kempian

Kempian's AI Features — Candidate Matching, Resume Parsing, the Job Creation Assistant, and the Chat Assistant — are the four AI components of the Platform. Each is described below, along with the data it processes and the level of impact it can have on a candidate's application journey.

2.1 Candidate Matching

What it does. Candidate Matching uses large language models (LLMs) and embedding-based similarity techniques to score and rank candidates against the requirements of a job posting. The output is a numeric or tiered match score, together with a short explanation of the factors that contributed to that score (see Section 5).

What data it touches. CV/resume content, employment history, skills, location, and related profile data supplied by or about the candidate.

Decision impact. High. Match scores directly inform which candidates a recruiter is likely to look at first, so this is the AI component with the greatest potential influence on a candidate's visibility in a hiring process. For that reason, match scores are recommendations only — see Sections 3 and 4.

External providers. OpenAI, Anthropic, and HuggingFace (see Section 8).

2.2 Resume Parsing

What it does. Resume Parsing extracts structured data — such as job titles, employers, dates, education, and skills — from an uploaded CV or resume, so that it can populate or update a candidate's profile.

What data it touches. The full content of the submitted CV or resume, which may include sensitive career details.

Decision impact. Medium-high. Parsing errors can misrepresent a candidate's background in ways that affect how they are matched and how they appear to recruiters. Candidates and recruiters should actively review extracted profile data for accuracy — see Section 3 for more on how this works today.

External providers. OpenAI, together with an internal/locally hosted model for parts of the extraction pipeline.

2.3 Job Creation Assistant

What it does. The Job Creation Assistant generates draft job description text based on inputs an employer or recruiter provides (role title, requirements, seniority, etc.). It is a drafting aid, not a publishing tool — the output is a starting point intended for editing.

What data it touches. Job requirement inputs provided by the employer/recruiter. This component does not process candidate personal data.

Decision impact. Low. It does not affect any individual candidate's treatment directly, though a published job description can indirectly shape who applies. Employers remain responsible for reviewing generated text before it is posted (see Section 7 on hallucination risk).

External providers. OpenAI.

2.4 Chat Assistant

What it does. An in-Platform conversational assistant that answers user questions about the Platform, a role, or an application, and can help users navigate Kempian's features.

What data it touches. Conversation content entered by the user, which may reference profile or CV data already held on the Platform.

Decision impact. Low, in the sense that the Chat Assistant does not itself rank, screen, or select candidates. As with any generative AI text, its responses should be treated as a starting point for review rather than an authoritative answer (see Section 7).

External providers. Internal backend systems (no named third-party model provider for this component at this time).

3. Where Humans Remain Responsible

AI does not make the final hiring, rejection, or shortlisting decision on Kempian. Every AI output capable of affecting a candidate — a match ranking, an outreach message, a resume-derived profile field, or a generated job description — is designed to pass through a human recruiter or employer before it takes effect. Recruiters and employers can review, edit, and override AI outputs, and remain responsible for the hiring decisions they make using the Platform.

This is Kempian's Human Review Gate: the principle that every AI Feature output capable of affecting a candidate requires human review before it takes effect. Kempian is implementing human-confirmation checkpoints across its AI-assisted profile updates as part of its ongoing compliance program. In the meantime, candidates and recruiters should review AI-derived profile data — including resume-parsed fields — and correct anything that does not accurately reflect a candidate's actual background.

If you believe an AI output has affected you without appropriate human review, you can request review using the process in Section 10. Enterprise customers seeking full technical detail on the current implementation status of the Human Review Gate can request the Kempian AI Governance Statement.

4. Confidence Scores and Recommendations

Match scores are generated by comparing a candidate's profile data against the stated requirements of a job, using a combination of LLM-based reasoning and embedding similarity. The resulting score or ranking is a relevance indicator, not a hiring decision. It reflects how closely the system judges a candidate's stated qualifications align with a role's stated requirements, based on the data available to it at the time of scoring.

A confidence indicator accompanies each score where available. It reflects how much weight the system itself places on the result. A lower confidence indicator signals that the underlying CV or job data was sparse, ambiguous, or inconsistent, and that the score should be weighted accordingly during human review. A higher confidence indicator means the system found clearer, more consistent signal in the data it processed. It does not mean the score is guaranteed to be accurate.

Every match score is shown together with an explanation of the factors that contributed to it (Section 5), so a recruiter can assess it rather than accept it at face value. A recruiter's own review and decision always control. The Platform does not act on a match score or ranking independently of that review, consistent with the Human Review Gate described in Section 3.

Match scores should not be treated as a certification of a candidate's suitability. A low score does not mean a candidate is unqualified — it means the system did not find strong alignment based on the inputs it processed, which themselves can be incomplete or imperfect (see Section 6).

5. Explainability — "Why This Match?"

Alongside a match score, Kempian shows a short, plain-language explanation of the factors that contributed to that score. This explanation is designed to cover two things:

• Contributing factors — the elements of a candidate's profile that align with the role, such as overlapping skills, relevant experience, location fit, or seniority alignment.
• Missing or unclear qualifications — where relevant, the explanation can also identify requirements stated in the job posting that the candidate's profile does not clearly demonstrate, so a recruiter can see what is absent from a profile relative to the role, not only what matched.

This "Why this match?" explanation is designed to help recruiters and, where shown, candidates understand the basis for a score rather than treat it as a black box, consistent with explainable AI principles. It reflects the factors the system identified as influential. It is a summary intended to aid human judgment, not an exhaustive or certified account of every input that affected the underlying model's output.

Recruiters should use the explanation as one input into their own review, not as a substitute for reading a candidate's actual profile or CV. Where a recruiter's own assessment differs from the explanation — for example, because the recruiter has context the system did not have — the recruiter's judgment governs.

6. Limitations of the AI Systems

No AI system used on Kempian is error-free, bias-free, or guaranteed to be fully accurate. None of the AI components described in this notice should be relied on without human review. Known categories of limitation include:

• Variable accuracy depending on CV quality and format. Resume parsing and downstream matching accuracy depend on how a CV is formatted, structured, and written. Unusual formats, scanned documents, or non-standard layouts are more likely to produce extraction errors.
• Extraction errors. Automated extraction can misread, omit, or misattribute information — for example, mismatching dates to roles, missing a certification, or misinterpreting an abbreviation. This is why candidates and recruiters are asked to check parsed data (Section 3).
• Score sensitivity to data quality. A match score is only as good as the underlying profile and job data. Incomplete profiles, outdated CVs, or vaguely written job requirements can all produce less reliable scores.
• Non-determinism of LLM outputs. Large language models can produce different outputs for similar or even identical inputs, and outputs can change as models are updated over time. A score, explanation, or generated text should not be assumed to be perfectly reproducible.
• Bias and fairness testing is an ongoing program, not a completed guarantee. Kempian is building a program of testing designed to identify and reduce disparate outcomes across protected groups in AI-assisted matching and screening. This work is in progress. Kempian does not claim its systems are "bias-free" — no organization using AI in hiring honestly can. Testing methodology, scope, and independent audit status are described further in the Kempian AI Governance Statement.

7. Hallucination Risk in Generative AI Outputs

Two of Kempian's AI components — the Job Creation Assistant and the Chat Assistant — generate new text rather than only extracting or scoring existing data. Generative AI systems of this kind can produce content that sounds plausible but is inaccurate, incomplete, or entirely fabricated — a phenomenon commonly called "hallucination." This can include, for example, a drafted job description that states a requirement, benefit, or fact that was not provided as input, or a chat response that answers confidently but incorrectly.

All AI-generated text on Kempian is provided for human review before use, not for direct reliance. Employers and recruiters should read and edit any AI-drafted job description before publishing it. They should not treat Chat Assistant responses as a definitive or binding statement of fact about a role, a policy, or an application's status. If you are unsure whether AI-generated text is accurate, verify it against the underlying source (the actual job requirements, your actual profile, or a human recruiter) before acting on it.

8. Third-Party AI Providers

Kempian uses the following named third-party AI providers within the components described in Section 2: OpenAI, Anthropic, and HuggingFace. Kempian also uses internally hosted/backend models for parts of resume parsing and the Chat Assistant, as noted above.

Kempian is the deployer of these third-party models, not their developer. Kempian remains legally responsible for how their outputs are used on the Platform, regardless of which provider generated a given output. Using a third-party AI provider does not shift that responsibility to the provider.

Kempian requires data-handling commitments from each provider by contract, consistent with industry-standard enterprise/API terms. In particular, data submitted through Kempian is not used to train the provider's general-purpose models, and is subject to contractually defined retention limits. Kempian is finalizing formal Data Processing Agreements with these providers as part of its ongoing vendor management program. Enterprise customers can request current status detail through the Kempian AI Governance Statement.

9. User Responsibilities

Because AI outputs on Kempian are recommendations and drafts rather than final decisions, candidates and recruiters share responsibility for using them appropriately:

• Review AI outputs critically rather than treating them as automatically correct or authoritative.
• Check resume-parsed profile data for accuracy, and correct any field that misrepresents your background (Section 3).
• Do not rely on AI-generated job descriptions or chat responses as a final or binding statement without human verification (Section 7).
• Report suspected errors, unexpected AI behavior, or output that appears biased or discriminatory using the process in Section 10.

Additional obligations on how AI Features may and may not be used are set out in the Kempian AI Acceptable Use Policy.

10. Requesting Human Review or Objecting to AI-Driven Processing

If you are a candidate and believe an AI output has affected your application without adequate human review, or you wish to object to a particular use of AI in how your data was processed, you may contact Kempian at privacy@kempian.com. Depending on your jurisdiction, you may have a right to obtain human intervention, to express your point of view, and to contest a decision that was based solely on automated processing. Kempian's Privacy Policy describes these rights in full for your applicable jurisdiction.

Employers and recruiters who have questions about a specific AI output, or who wish to report a suspected malfunction or bias concern, should use the same contact channel or their designated Kempian account contact.

11. Regulatory Context

Kempian's recruitment-related AI functionality — candidate matching and resume screening in particular — falls within the category of high-risk AI systems used in employment under Annex III of the EU AI Act. Transparency obligations requiring disclosure of AI use to affected individuals (Article 50) are already in effect.

Broader high-risk obligations, such as conformity assessment and formal technical documentation, apply on a later timeline under current EU rules. Kempian is working toward those obligations ahead of the applicable deadline (see the Kempian AI Governance Statement for detail). Additional disclosure or notice obligations may apply depending on the jurisdiction of the employer, recruiter, or candidate involved, including under US state and India law referenced in Kempian's Privacy Policy.

This notice does not constitute legal advice to any user and should not be relied upon as a complete statement of applicable law in every jurisdiction.

Related Documents

This notice should be read together with: the Privacy Policy, the Terms and Conditions, the AI Acceptable Use Policy, the AI Governance Statement, and the Candidate Privacy & Visibility Notice.

12. Contact

Questions about this notice, or about how AI is used on Kempian, can be directed to privacy@kempian.com, attention: DPO name/contact — to be appointed.

Kempian — AI Transparency Notice — v0.3 (Draft) — July 2026