Responsible AI training module

Purpose

This is the required responsible-AI training module for HiringCoachAI personnel. HiringCoachAI uses AI as a core product capability, so the goal is not to avoid AI. The goal is to use approved AI features and vendors in a way that protects users, avoids overclaiming, and treats model output as assistive.

Complete this module before working on AI-assisted product features, reviewing AI outputs, handling AI-related support issues, or receiving sensitive operational access.

1. Use Approved AI Systems Only

Production customer data may be sent only to approved AI providers and only through approved product paths or approved operational procedures. Do not paste customer data, logs, resumes, transcripts, or screenshots into personal AI accounts or unapproved tools.

Approved AI vendors and feature mappings are maintained in AI model inventory and AI use disclosure.

2. Minimize What The Model Sees

Send only the data needed for the requested feature. Avoid including unrelated profile data, account details, contact data, support history, secrets, or logs in prompts.

Never send Restricted data to an AI model, including API keys, service-account keys, OAuth tokens, session cookies, backup keys, or payment card data.

3. Treat AI Output As Assistive

AI output is not final truth. Resume, cover-letter, pitch, and similar drafting outputs are intended for user review and editing. Scoring, transcription, task breakdowns, and value-proposition suggestions are informational aids.

HiringCoachAI does not use AI to make automated employment decisions with legal or similarly significant effects.

4. Watch For Bias, Hallucination, And Unsafe Output

Report AI behavior that appears:

• Biased or demographically suggestive.
• Materially wrong or fabricated.
• Privacy-invasive or overbroad.
• Unsafe, harassing, deceptive, or discriminatory.
• Misleading about a user's qualifications or employment prospects.

Reports feed prompt review, model inventory updates, and the AI bias evaluation (annual cadence intended; first annual cycle in progress).

5. Preserve User Notice

HiringCoachAI gives layered notice through the first-visit banner, the public privacy policy, and the AI disclosure page. Do not ship a new AI feature, new AI provider, or materially different trigger without updating the AI disclosure and model inventory.

Some features run when a user enters data into an AI-assisted flow, such as job-description data triggering Fit Analysis. That behavior must be described plainly in the disclosure.

6. Respect Retention And Logging Limits

Where supported, HiringCoachAI passes provider controls to reduce storage or transcript exposure, such as store: false for OpenAI and redact=true for Deepgram transcripts. Provider retention posture is documented in the AI use policy and model inventory.

Internal AI audit logs should record metadata such as model, endpoint, timing, and token counts. They should not record prompt text or model response content unless a future approved control explicitly permits it.

Acknowledgment

By acknowledging this module in /admin/onboarding, I confirm that I understand HiringCoachAI responsible-AI expectations, will use only approved AI systems for customer data, will treat outputs as assistive, and will report biased, unsafe, privacy-invasive, or materially wrong AI behavior.