AI System Disclaimer

1. How AI Scoring Works

sHRark uses Azure OpenAI (GPT-5.4) to analyse candidate CVs and score them against a structured role brief you define. The system evaluates four dimensions: skills match, culture fit, growth potential, and risk indicators.

Scores are composite values in the range 0–100 derived from:

• CV content analysed against your role requirements and context.
• Feedback and interview notes entered by your team.
• Aggregated signals from similar candidates in your pipeline.
• Reference CVs you provide as anchors (if configured).

2. Limitations and Known Risks

• AI errors: The model may misread CV content, misunderstand role requirements, or produce inconsistent results across similar candidates.
• Bias risk: AI models can reflect biases present in training data. sHRark includes deterministic fairness guards (skills floor caps, evidence-first ranking, style-blind CV processing) but these do not eliminate bias entirely.
• CV quality dependency: Scores depend on the quality of CV text extracted by OCR and parsing. Poorly formatted CVs may produce lower-quality scores.
• Language: The system is optimised for Polish and English. CVs in other languages may produce less reliable results.
• Not a predictor of job performance: AI fit scores are not validated predictors of future job performance or employee success.

3. Human Oversight Requirement

Under the EU AI Act, AI systems used in employment contexts are classified as high-risk. sHRark is designed as a human-in-the-loop tool. You must:

• Review AI scores alongside the original CV and your own assessment.
• Not advance or reject candidates based on AI score alone.
• Disclose AI involvement in screening to candidates upon request.
• Maintain an audit log of hiring decisions (available via sHRark export).

4. Fairness Controls

sHRark includes several deterministic, non-AI fairness layers:

• Style-blind CV processing: summary sections stripped before scoring.
• Evidence-first guard: summary-heavy CVs are capped and flagged.
• Skills floor cap: culture fit cannot compensate for low skills score.
• Minimum requirement gate: candidates failing hard requirements are capped.
• Rerank bound: comparative re-ranking is limited to ±15 points.

These controls reduce but do not eliminate the risk of unfair outcomes. Regular audits of scoring outcomes across demographic groups are recommended.

5. Data Used for Scoring

Scoring uses only data your organisation has uploaded or entered: CVs, role briefs, and interview notes. Candidate data is not shared between organisations and is not used to train AI models.

6. Compliance and Responsibility

sHRark is provided as a decision-support system. The customer is responsible for ensuring that use of the system within its hiring process complies with applicable privacy, employment, anti-discrimination, workplace, and AI-governance laws, including any required notices, human review, or internal approval procedures.

Nothing in this page should be understood as legal advice or as a guarantee that the customer's use of AI in employment is compliant without its own review.

7. Contact

Questions about AI scoring or fairness? ai@shrark.com