Our Responsible AI Commitment
Effective Date: 28 May 2026 Version: 1.0 Owner: Altara AI Governance Office Issued by: Altara AI · Part of the Navigate Compliance Group (Pty) Ltd
Altara AI is built on a single conviction: artificial intelligence must earn the trust of the people it serves. Responsible AI is not a label we use — it is the operating principle that shapes every model, every workflow, every release decision, and every line of policy at Altara.
This page sets out the public commitments we make to our users, our clients, our regulators, and the communities our platforms touch.
1. Our Six Commitments
1.1 Human Accountability
A human is always accountable for any decision informed by an Altara AI system. AI augments judgment; it does not replace it. Our platforms are designed so that high-impact outputs are reviewable, challengeable, and overrideable — and so that the human accountable for each decision is identifiable.
1.2 Transparency
We disclose, in plain language, when our systems use AI. Every NAVI-generated output carries a model version, a prompt version, and a confidence rating. The indicator catalogues, the scoring logic, and the model cards behind our risk assessments are published — not hidden behind marketing copy.
1.3 Explainability
Every Altara AI verdict is explainable. We don't ship opaque scores. When NAVI flags a piece of content as a likely scam, the user sees the indicators that triggered, the verbatim evidence from the submission, and the reasoning in plain language. Regulators can audit the methodology, not just the output.
1.4 Fairness
We actively test our systems for bias — across languages, demographics, regions, and use cases. We document known limitations. We commit to remediating disparate impact when it is identified, and to disclosing limitations to users before they rely on a result.
1.5 Safety and Robustness
We monitor our AI systems continuously for hallucinations, adversarial manipulation, performance degradation, and emerging threats. Our incident-response playbooks include rapid model rollback, indicator-catalogue freeze, and public disclosure where consumer harm is plausible.
1.6 Privacy by Design
We collect the minimum information needed to do the job. Identifiers are not requested where they are not necessary. Data residency, encryption, retention limits, and user-rights controls are designed into our systems — not bolted on. See our Privacy Policy for the full detail.
2. The Frameworks We Align With
Our Responsible AI programme is informed by — and continuously cross-walked against — the leading global frameworks:
| Framework | What it gives us |
|---|---|
| OECD AI Principles | The international baseline for trustworthy AI |
| NIST AI Risk Management Framework (AI RMF 1.0) | Concrete control families for govern / map / measure / manage |
| ISO/IEC 42001 | A certifiable AI management system standard |
| ISO/IEC 23894 | AI risk-management guidance |
| EU AI Act | Risk-tiered regulatory expectations (general-purpose, limited, high-risk, prohibited) |
| POPIA & GDPR | Lawful, fair, and minimal processing of personal information |
| Responsible AI best practices | Industry-developed norms (e.g. Microsoft RAI standard, Partnership on AI, Singapore Model AI Governance) |
We track changes to each of these continuously and update our internal control mapping when a framework evolves.
3. How We Operationalise Responsible AI
| Practice | Where you can see it |
|---|---|
| Human-in-the-Loop on every consequential output | HITL Governance Policy |
| Published model cards | NAVI (Claude Sonnet 4.5) and Sentinel Transcribe (OpenAI Whisper) cards — covering training-data lineage, intended use, out-of-scope use, evaluation dimensions, and known limitations |
| Indicator catalogue is public | The 30+ scam indicators used by NAVI Trust Intelligence are visible to anyone who interacts with the platform |
| Versioned scoring + prompts | Every output is stamped with the scoring-model version and prompt version. Regulators can interrogate any past assessment |
| Audit trails | User interactions, AI-generated outputs, human approvals, overrides, escalations and monitoring logs are retained where feasible to support accountability |
| Privacy-aligned data handling | See our Privacy Policy — submitted content is not linked to identifiable users in anonymous-by-default surfaces like Trust Lens |
| Disclaimers on every output | Every NAVI output includes "AI-assisted assessment — not a final determination of fraud. Verify with the relevant regulator before acting." |
| No surveillance, no scraping, no dark patterns | We earn data through explicit opt-in submissions only |
4. What Responsible AI Means For Different Stakeholders
For users and the public
You will always be told when you are interacting with AI. You will see what the AI found, why it found it, and where it might be wrong. You can withdraw consent and erase your on-device data at any time on apps that store it locally. You will never be presented with an AI output as if it were a human professional opinion.
For institutional clients (banks, regulators, ombud schemes, fintechs)
You receive a governance-grade product — explainable scoring, audit trails, HITL workflows, override mechanisms, and the right to challenge any output. Our platform is built so that your compliance team can defend every AI-informed decision in front of a supervisor.
For regulators
We publish methodology, model cards, indicator catalogues and our HITL Governance Policy proactively. We participate in the Global Regulatory TechSprint and engage with national supervisors on consumer-protection and trust-intelligence use cases. Our systems are designed to be interrogable by a non-technical regulator, not just by an ML engineer.
For our employees, contractors and partners
Our internal AI Governance Office reviews every new model, prompt, and indicator change. Responsible AI training is mandatory and recurring. Whistle-blower channels are open for raising concerns about AI behaviour, bias, or misuse without retaliation.
5. What We Will Not Do
There are things responsible operators of AI must refuse. Altara AI commits that we will not:
- Develop, sell, or operate AI systems for social scoring of natural persons.
- Develop AI for biometric mass surveillance of the public.
- Develop AI that manipulates individuals through deceptive techniques into decisions that harm them.
- Sell user content (including the text people paste into our analysers) to data brokers.
- Use submitted content to train third-party generative models without explicit, separately-collected consent.
- Misrepresent AI-generated outputs as professional legal, financial or regulatory advice.
- Operate AI systems in jurisdictions where our use case is unlawful or unsafe.
These commitments are reviewed quarterly and may only be expanded — never narrowed.
6. Advocacy and Public Engagement
Altara AI actively advocates for responsible AI adoption across Africa and beyond:
- Public education — we publish open educational content on AI-enabled fraud patterns at our Insights blog and the Sentinel Investor Education library.
- Industry collaboration — we work with banks, regulators, ombud schemes and consumer-protection bodies to share threat intelligence and build community-facing defences.
- Policy contribution — we submit comments on AI regulatory consultations (e.g. South Africa's Information Regulator AI guidance, the FSCA's market-conduct AI principles).
- Open frameworks — where we develop reusable governance artefacts (indicator catalogues, model-card templates, HITL workflows), we publish them under permissive terms so other operators can learn from them.