Scovai operates an AI-powered recruitment platform that falls within the scope of the EU Artificial Intelligence Act (Regulation (EU) 2024/1689). AI systems used in recruitment and employment are classified as high-risk under Annex III, Section 4(a). This document describes how Scovai meets the requirements of the EU AI Act.
1. Risk Classification
The following Scovai AI features are classified as high-risk AI systems under Annex III point 4(a) of the EU AI Act. Features that evaluate people who are already employed are covered separately in section 8A.
- AI Candidate Scoring: Automated scoring of candidates across multiple dimensions (technical skills, experience, education, soft skills, potential)
- AI Shortlist Generation: Automated ranking and recommendation of candidates for positions
- Psychometric Profiling: AI-generated personality profiles (Big Five), culture fit assessments, and leadership style evaluations
- AI Interview Agent: Automated interview conduct and evaluation using natural language processing
The following features are considered limited-risk and are subject to transparency obligations only:
- CV Parsing: Extraction of structured data from uploaded documents
- Job Description Generation: AI-assisted creation of job descriptions from minimal input
- Technical Assessment Generation: AI-generated role-specific test questions
2. Transparency (Article 13)
Scovai provides transparency at every level of AI processing:
2.1 Interaction Transparency
Candidates are informed when they are interacting with an AI system. The AI Interview Agent clearly identifies itself as an AI, and all AI-generated content (scores, assessments, reports) is labelled as such throughout the Platform.
2.2 Output Explainability
Every AI output includes human-readable explanations:
- Scoring: Each score dimension includes an XAI rationale explaining which factors contributed, what adjustments were made, and why
- Psychometric profiles: Results include descriptions of what each trait means and how it was assessed
- Interview reports: AI evaluations cite specific candidate responses and explain the basis for each assessment
- Shortlists: Ranking rationale explains the weight distribution and key differentiators between candidates
2.3 System Documentation
We maintain comprehensive technical documentation of our AI systems, including their intended purpose, capabilities, limitations, and known risks.
3. Human Oversight (Article 14)
Scovai is designed as a decision-support system, not a decision-making system. Human oversight is embedded in the platform architecture:
- No autonomous hiring decisions: AI outputs are presented as recommendations. Human recruiters review, validate, and make all final decisions.
- Override capability: Recruiters can override, adjust, or discard any AI-generated score, ranking, or recommendation.
- Human review requests: Both candidates and recruiters can request formal human review of any AI decision through the compliance module. Requests are tracked and audited.
- AI feature controls: Tenant administrators can enable or disable individual AI features at any time.
- Configurable weights: Scoring weights across all dimensions are configurable by the organisation, ensuring AI behaviour aligns with human-defined priorities.
4. Data Governance (Article 10)
Our AI systems are governed by strict data quality and governance practices:
- Training data: We do not use individual candidate data to train or fine-tune AI models. Our AI providers process API data under data-processing agreements that exclude its use for training their models.
- Data quality: CV parsing includes validation checks. Scoring inputs are verified against structured position requirements.
- Data minimisation: AI systems only receive the data necessary for their specific task.
5. Accuracy, Robustness, and Cybersecurity (Article 15)
- Accuracy: AI scoring is validated against structured requirements with configurable thresholds. Explainability rationale allows recruiters to verify the basis of each score.
- Robustness: AI systems include timeout protections, retry logic, and fallback routing to alternative models if the primary model is unavailable.
- Cybersecurity: The platform and its database run on our own servers in the EEA; Digital Employee runtimes run on our own servers outside it, as described in the Privacy Policy and the DPA. AI model calls are routed through centrally configured, audited providers (Anthropic, OpenAI) under data-processing agreements. API keys are encrypted at rest with AES-256 (CBC for provider keys, GCM for Digital Employee credentials), and all communications use TLS 1.2 or 1.3.
6. Bias and Discrimination Prevention (Article 10(2)(f))
Scovai implements proactive bias monitoring and mitigation:
- Automatic bias analysis: The analytics module includes bias monitoring that tracks AI scoring patterns across demographic groups (gender, age)
- Score distribution analysis: Recruiters can view score distributions to identify potential systematic biases
- Configurable scoring: Scoring dimensions and weights are transparent and adjustable, preventing hidden bias amplification
- Explainability: By providing clear rationale for every score, biased reasoning can be identified and addressed
This monitoring covers the recruiting funnel. It does not currently extend to Digital Employee output, performance ratings or culture assessments.
7. Record-Keeping and Audit Trail (Article 12)
Scovai maintains comprehensive logs and records:
- Audit log: Every platform action (score generated, candidate assessed, shortlist created, decision made) is recorded with actor, timestamp, and details
- Structured logging: System logs capture AI processing events with severity levels, categories, trace IDs, and duration metrics
- Retention: Critical and error logs are retained for 365 days in the live system, and standard logs for 30-90 days depending on severity. Each row is written to a compressed archive file before it is deleted from the live system; those archive files are kept until an administrator removes them.
- Export: Audit data can be exported for regulatory review
8. Registration and Conformity
As a provider and, for our own use, a deployer of high-risk AI systems in the employment domain, our obligations are:
- Registration of high-risk AI systems in the EU database under Article 49. This registration has not yet been made. We will record its status here when it is.
- Maintaining conformity assessment documentation
- Cooperating with national competent authorities and market surveillance bodies
- Reporting serious incidents in accordance with Article 73
Two further obligations fall on the Tenant as deployer, and we state them here rather than leave them implied. Article 26 requires human oversight by people with the competence to exercise it, and Article 4 requires that staff who use these systems have sufficient AI literacy. Where the deployer is a body governed by public law, or assesses creditworthiness or life and health insurance pricing, Article 27 also requires a fundamental rights impact assessment. We do not currently provide templates for either; Section 5.2 of the Companies Terms sets out the division of responsibility.
8A. Digital Employees and Employment Features
A Digital Employee that screens, ranks or interviews candidates is a high-risk AI system under Annex III point 4(a) of the Act, on the same footing as the recruiting features in section 1, and is covered by sections 2 to 7 of this page.
Scovai also runs AI on people who are already employed: performance and year-end review synthesis, development plans, culture and onboarding assessments, check-in summaries, and competency scorecards. Annex III point 4(b) covers AI intended to be used for decisions on promotion or termination, for the allocation of tasks based on behaviour or personal traits, and for monitoring and evaluating performance. Where a Tenant uses these outputs for such decisions, they fall within that point, and our Terms already state that performance evaluation and task allocation are high-risk uses.
In the platform these outputs are advisory. The numbers in a competency scorecard are computed by a deterministic formula, not by a model; the model writes only the narrative layer, and when it fails the scorecard ships with the numbers alone. Integrity signals never reject a candidate on their own. The Tenant, as deployer, remains responsible for human oversight, for informing workers and their representatives, and for any impact assessment its own national law requires.
Digital Employees are marked as AI inside Scovai and introduce themselves as AI at first contact; the Terms describe how they identify themselves to third parties.
9. Candidate Rights Under the EU AI Act
As an individual affected by Scovai's AI systems, you have the right to:
- Be informed that an AI system is being used in the recruitment process
- Receive an explanation of how AI assessments, scores, and rankings were generated
- Request human review of any AI-generated decision
- Lodge a complaint with the relevant national authority if you believe the AI system has been used in a non-compliant manner
10. Contact
For questions about our EU AI Act compliance or to report concerns about AI decision-making:
Email: [email protected]
