| Failure Mode | Description | Mitigation Strategy |
|---|---|---|
| 1. Data Bias | AI learns patterns from skewed historical data (e.g., over-representation of certain education paths or demographics). | - Conduct periodic retraining with fresh, diverse data. - Include underrepresented profiles in training datasets. - Run External Bias Audit every 6 months |
| 2. False Positives/Negatives | Candidates are incorrectly scored too high or too low. | - Set retake limits per job - Require human review of borderline candidates. - Calibrate criteria to ensure accuracy |
| 3. Gaming the System | Candidates use prep tools or copy scripts from online forums to "beat" the interview. | - Encourage human reviewers to flag suspect responses. |
| 4. Privacy Risks | Candidate data is mishandled, shared improperly, or retained beyond legal limits. | - Enforce secure infrastructure (AES-256, TLS 1.2+). - Follow data minimization and retention policies (see Data Management Policy). - Obtain explicit, revocable candidate consent. |