AceHR · Legal & privacy
AI Transparency & Automated Decision-Making Notice
How AceHR uses AI in recruiting and assessments, what goes in and comes out, where people review it, what AI never does, and how candidates can ask for human review.
Last updated:
1. Our approach
AceHR uses AI to help recruiters and hiring teams work faster and with better evidence. It does not make hiring decisions. The staffing firm or employer using AceHR (the “customer” or “hiring company”) decides whether to use each AI feature, configures it, and remains responsible for every employment decision. Under laws that regulate AI in employment, the customer is usually the employer or deployer using the tool, and AceHR is the developer or vendor that supports it.
We design AI features around four commitments: outputs are advisory; analysis cites the evidence it relies on; consequential steps require human review or approval; and candidates can ask for a person instead.
2. AI features at a glance
The table below lists the AI and automated features in AceHR, what goes in, what comes out, and where a person reviews the result. Customers choose which features to use.
| Feature | Inputs | Outputs | Human review |
|---|---|---|---|
| Résumé parsing and evidence extraction | Résumé file or text; job description. | Structured profile, skills and experience, with cited source passages. | Recruiters see the source passages and can correct records. |
| Requirement matching and scoring | Job requirements, extracted evidence and rule-based experience calculations. | A score computed by versioned rules from the evidence, and a separate advisory AI opinion with citations. | Recruiters review and decide. Customers can reduce or turn off AI influence on scoring. No automatic rejection. |
| Summaries, reports and drafts | Workspace records the user selects. | Candidate summaries, client reports, and email and message drafts. | A user edits and chooses whether to send or share. |
| Interview and application question drafting | Job description and requirements. | Draft questions that cite the job text; optional knockout criteria with a job-related rationale. | A recruiter must review and publish. Unmet knockout criteria are flagged for review, never auto-rejected. |
| AI voice and browser interviews | Approved questions, job context and the candidate’s spoken answers. | Transcript, per-question evidence and scores, and noted inconsistencies to clarify. | A recruiter reviews transcripts and recordings and records the decision. Candidates who ask for a human are not scored. |
| Assessment generation | Job description, the recruiter’s brief and a summary of the candidate’s résumé skills and experience. | Draft tasks, tests and private rubrics. | A recruiter edits and approves before inviting the candidate. |
| Coding hints assistant (optional) | The task’s public prompt and samples, the candidate’s current code and question. | Concepts and debugging hints, within a set number of requests. | The conversation is saved for the reviewer. Approved use is not penalized. |
| Final explanation questions | The candidate’s submitted work and the task. | Questions about the candidate’s own approach (standard questions if AI is not used). | Shown to the candidate; the answers are reviewed by a person. |
| Transcription and score suggestions | Recorded explanation audio, submitted answers, test results and the private rubric. | Transcript, suggested task scores with cited evidence, and an advisory recommendation. | A reviewer scores every task and makes the decision. |
| Integrity risk indicator | Monitoring observations, identity check status and network indicators. | Low, Medium or High review priority, with reasons. | Computed by fixed rules, not AI. A reviewer records a separate integrity judgment. |
| On-device vision | Camera frames, processed only in the candidate’s browser. | Observations such as face not visible, multiple faces, gaze away, or a phone or person in view. | Reviewed alongside the recording. The detector’s measurements are never stored. |
| Search | Job and candidate text converted to embeddings. | Relevance ranking for recruiter searches. | Users choose which results to open and act on. |
| Assistant, copilot and workflow agents | Workspace information the user can already access, and the user’s request. | Answers, analyses and proposed actions. | Agents propose; a person approves actions such as scheduling. |
| Job seeker career comparison | A signed-in job seeker’s own career profile and a public job. | A comparison of stated skills and experience with the job’s requirements. | For the job seeker only; not used in any employer’s decision. |
3. What AceHR’s AI does not do
By design, AceHR’s AI features:
- do not reject, advance, hire or place anyone on their own; screening flags, scores and indicators require human review;
- do not use facial recognition to identify people; the only face matching is the one-to-one identity check that our identity verification provider (Persona, or Stripe where an assessment uses it) performs when a hiring company requires it;
- do not analyze facial expressions, emotion, personality, honesty, accent, speaking style or appearance, and do not infer them from typing or voice;
- are not designed to use protected characteristics such as race, sex, age, disability, religion or national origin; our prompts instruct models not to infer or use them or proxies such as names, photos, addresses or graduation years, although documents can still contain such information, which is one reason human review is required;
- do not label candidates dishonest or treat polished writing as evidence of AI use; and
- do not produce consumer reports or background checks.
4. Known limitations and how we manage risk
AI can be wrong. It can misread unusual résumé layouts, miss or misstate facts, reflect biases in the data it was trained on, be influenced by text hidden in documents, and give different answers to the same input. Speech-to-text can be less accurate for some accents, speech differences or poor audio. On-device vision can be affected by lighting, glasses, camera angle, skin tone, disability and shared spaces. Network indicators can flag ordinary VPNs and workplace networks.
We manage these risks by computing scores with deterministic, versioned rules and keeping AI opinions separate; requiring citations to source evidence and checking that cited evidence exists; versioning prompts and recording which prompt and model produced an output; treating uploaded content as untrusted data to reduce prompt injection; instructing models to avoid protected characteristics; capping and explaining the points behind integrity indicators; giving reviewers the underlying evidence; and monitoring errors and usage. These measures reduce but do not eliminate risk, and they are not a bias audit.
5. Customer responsibilities under AI employment laws
Customers decide where and how these features are used, so they are responsible for the laws that apply to their hiring, including:
- New York City Local Law 144: an independent bias audit within one year before using an automated employment decision tool, a published summary of the results, notices to candidates and employees at least 10 business days before use, and information on requesting an alternative process or accommodation.
- Illinois: notifying applicants and employees when AI is used in covered employment decisions, and avoiding discriminatory effects, under the Illinois Human Rights Act from January 1, 2026; and, for Illinois-based positions, the Artificial Intelligence Video Interview Act’s notice, explanation, consent, sharing and deletion rules.
- Colorado: deployer duties for high-risk AI systems under the Colorado Artificial Intelligence Act, when and to the extent it is in effect.
- California: the Civil Rights Council’s automated-decision system regulations under the Fair Employment and Housing Act, and, from January 1, 2027, the CCPA regulations on automated decision-making technology where they apply.
- Maryland: a signed applicant waiver before facial recognition is used during an interview.
- Federal and state anti-discrimination laws: Title VII, the Americans with Disabilities Act and the Age Discrimination in Employment Act apply to AI-assisted decisions, including the duty to provide reasonable accommodations.
AceHR helps by providing this notice, built-in candidate notices and consent steps, human review workflows, evidence and audit records, and, on request, the information about our features that customers reasonably need for bias audits and impact assessments. AceHR does not perform bias audits for customers or certify that any use is lawful.
6. Statement for Colorado and similar laws
To the extent any AceHR feature is a high-risk AI system under the Colorado Artificial Intelligence Act or a similar law, this notice is our public statement of the types of systems we offer (candidate analysis and matching, and assessment and interview evaluation aids) and of how we manage known or reasonably foreseeable risks of algorithmic discrimination, as described in “Known limitations and how we manage risk”. Customers can request further documentation about intended uses, data, limitations and evaluation using the contact details below.
7. Human review, alternatives, correction and appeal
If you are a candidate, you can:
- ask the hiring company to have a person review any AI-assisted result about you;
- request a human alternative or accommodation in assessments and AI interviews at any time, without an automatic penalty;
- tell the hiring company if your résumé, profile or other information was read incorrectly, and ask it to correct the record and reconsider;
- appeal a decision where the law or the hiring company’s process provides an appeal, by asking the hiring company; and
- contact AceHR at acehrassociatesinc@gmail.com if you cannot reach the hiring company, with its name and the job, and we will pass your request to it.
AceHR does not make hiring decisions and cannot change a hiring company’s decision, but we help customers respond by providing the underlying evidence and records.
8. AI model providers
We use models from Anthropic (Claude, our primary provider), OpenAI (including speech-to-text and embeddings) and Google (Gemini) through their commercial APIs, and Retell AI powers AI voice interviews. AceHR chooses the provider for each task based on capability and availability. Providers process the data needed for each request under their agreements with us; see the subprocessor list. The on-device vision models are served from AceHR and run on the candidate’s device, so the detector does not send camera frames to the models’ publisher or any other provider.
9. Updates
We update this notice when we add or materially change AI features. The date above shows the latest version.