Artificial intelligence powers some of Slaeb’s most useful features — from anomaly detection in QHSE to the WhatsApp AI agent in CRM. We are committed to deploying AI ethically, transparently, and in a way that keeps humans firmly in control of the decisions that matter.
AI is one of the most powerful tools available to modern SMEs — but it is also one of the most easily misused. As a SaaS platform that builds AI directly into its modules, Slaeb takes a public position: artificial intelligence should make people’s working lives better, not replace their judgment, surveil their behaviour, or hide decisions behind a black box.
This Responsible AI Policy sets out how we design, train, deploy, monitor and improve AI features across Slaeb. It complements our Privacy Policy and GDPR commitments, and is published so every customer knows exactly where we stand.
Every AI feature in Slaeb is designed and reviewed against five non-negotiable principles:
AI outputs must not discriminate against any protected group. Models are tested for disparate impact before release.
Users always know when they are interacting with AI, what data informs the output, and how to override it.
Every consequential decision remains a human one. AI proposes; people decide.
AI features use the minimum personal data needed and process it under strict no-retention agreements.
Each model and prompt is tested for harmful, biased or hallucinated outputs before release and continuously after.
We prefer efficient, small models where possible to limit energy use and environmental impact.
AI is not used everywhere. We add it only where it delivers real value to users — and we tell you exactly where.
Qualifies inbound leads using a SPIN-based conversational approach, updates the pipeline and books meetings. Operates only on conversations explicitly initiated by leads.
Risk level:
Limited
Human override:
Always available
Provider:
Anthropic
Spots unusual patterns in incident reports, audit results or compliance indicators and alerts the QHSE manager. Does not auto-classify incidents.
Risk level:
Limited
Human override:
Always available
Provider:
In-house statistical model
Suggests optimal shift patterns based on historical data, leave requests and team availability. Suggestions only — managers always validate.
Risk level:
Limited
Human override:
Mandatory
Provider:
In-house model
Generates summaries of long documents (audit reports, contracts, CVs). Summaries are clearly labelled as AI-generated and always reviewed by a human.
Risk level:
Minimal
Human override:
Mandatory
Provider:
Anthropic
Natural-language search across your data ("show me last quarter's open audits"). Returns the actual records, not generated content.
Risk level:
Minimal
Human override:
N/A
Provider:
In-house embeddings
Where we don’t use AI — Slaeb never uses AI to make consequential decisions about people without human review. That means no automated hiring, firing, performance scoring, disciplinary action, or credit assessment.
AI features in Slaeb use only the data that is strictly necessary to deliver the feature. We work on the principle of data minimisation:
The single most important rule in this policy is that every consequential decision must remain a human one. AI in Slaeb is a copilot, never an autopilot.
No AI system is bias-free. Our approach is to acknowledge that, design against it, and continuously monitor for it.
| Stage | What we do |
|---|---|
Design |
Each feature has a documented “use case” and “out-of-scope” list, reviewed for potential bias |
Pre-release |
Red-team testing for discriminatory outputs across age, gender, ethnicity and disability |
Release |
Soft launch with opt-in beta users, qualitative feedback loop |
In production |
Continuous monitoring for output distribution shifts and user-reported issues |
Annual |
Independent audit of high-risk features |
You can tell when AI is being used in Slaeb because we always say so:
✦ AI labelAI processing of personal data is governed by GDPR like any other processing. Specifically:
Article 22 commitment — Slaeb does not currently deploy any feature that produces legal or similarly significant effects on individuals through solely automated means. If we ever did, we would explicitly notify affected users and offer human review on request.
We design Slaeb’s AI to meet — and where possible exceed — applicable AI regulations across our five markets:
| Region | Framework | Our alignment |
|---|---|---|
EU & France |
EU AI Act (Regulation 2024/1689) |
All current Slaeb features classified as Minimal or Limited risk; full risk register maintained |
United Kingdom |
UK Pro-Innovation AI Framework |
Aligned with the 5 cross-sectoral principles (safety, transparency, fairness, accountability, contestability) |
United States |
NIST AI Risk Management Framework + state laws |
Voluntary adoption of NIST AI RMF, monitoring of CA/NY/CO state-level AI legislation |
Caribbean (FR territories) |
Continuous monitoring for output distribution shifts and user-reported issues |
EU AI Act applies |
Asia (Singapore as anchor) |
Model AI Governance Framework (IMDA) |
Aligned with the eleven domains of the Framework |
If you believe an AI feature in Slaeb has produced a wrong, unfair or harmful output, we want to know.
We commit to acknowledging every report within 5 business days, investigating in good faith and, where appropriate, adjusting the feature.
As AI evolves and so does the regulatory landscape, this policy will evolve with it. Looking forward, we commit to:
No. Slaeb’s AI features provide suggestions and analysis, but every consequential decision — leave approval, performance review, hiring action — requires explicit human validation by a manager.
Yes. Workspace administrators can disable any AI feature at the module level. The rest of the platform continues to function normally.
Yes. All Slaeb features currently fall into the Minimal or Limited risk categories under the EU AI Act. We maintain a risk register and a documented compliance approach for each feature, and we will adjust as the Act’s provisions phase in through 2025–2027.
Through a four-stage process: pre-release red-team testing across protected characteristics, soft launches with feedback loops, continuous output monitoring in production, and annual independent audits for high-risk features.
No. Slaeb does not train, fine-tune or otherwise use customer data to improve foundation models. All third-party AI providers are bound by zero-retention contracts that purge data immediately after use.
We use Anthropic (Claude) for the WhatsApp AI agent and document summarisation, and in-house models for QHSE anomaly detection, HRMS scheduling suggestions and intelligent search. Standard Contractual Clauses and zero-data-retention agreements bind all providers.
Every AI output can be rejected, edited or reported via the “Report this output” button. We acknowledge every report within 5 business days, investigate, and adjust the feature where appropriate.
For any GDPR-related question, formal request or complaint.