Eswatini’s new AI Safety and Standards Centre of Excellence is proposing a route from institutional training to the technical work of assessing artificial intelligence. A statement distributed on 2 November by the Eswatini Standards Authority, ESWASA, and AICE Africa describes a phased programme, beginning with the standards body’s own operations.

The dates need care. Kenyan newspaper The Standard reported on 30 October that the launch took place on 27 October. The later statement sets out the programme in greater detail; it should not be read as evidence that a fully equipped testing laboratory opened this week.

For African businesses and public institutions buying AI systems, that distinction is consequential. A proposed centre, a trained workforce and an operational service capable of independently assessing products are different stages of development.

### Learning through implementation

The partners’ announcement, distributed through APO and reproduced by Africanews.it, describes four stages: internal transformation, practitioner development, industry services and regional cooperation. ESWASA provides institutional and standards expertise; AICE Africa contributes technical implementation and skills transfer.

One planned component is EswasaOne, a platform intended to connect standards information with certification and inspection workflows. The statement envisages a conversational interface and automation, accompanied by a framework setting responsibilities and data-governance requirements.

The proposed laboratory is a longer-term element. Certification services are also described as an intended development, dependent on validated methods, competent staff and arrangements for impartiality and accreditation. Those qualifications are central to the announcement.

### What counts as credible assurance?

The Standard’s reporting places the initiative within a wider concern about reliance on outside expertise. It identifies training for auditors, standards officers, regulators and technical personnel, alongside the proposed development of shared testing methods.

There is a useful distinction between possessing a tool and knowing how to challenge its outputs. A buyer considering an automated customer-service system, for example, needs to know how it performs with the languages, accents and questions its own customers use. Demonstrations using carefully selected examples cannot settle that question.

This is RIM’s analysis of the procurement task, rather than a claim about tests already performed by the centre. The appropriate evidence depends on the system and its intended use: recorded failures, reproducible assessment methods and clear responsibility for correcting problems.

### Standards need institutions behind them

ESWASA’s public explanation of its work describes standards as agreed documents setting requirements or guidance for products, services and processes. Its existing activities include training in standardisation and quality assurance.

That experience provides an institutional starting point. It does not establish that every existing certification service covers AI, or that a proposed AI assessment will automatically be accepted by buyers elsewhere. Users need the actual service scope and the evidence behind any assurance claim.

For creative organisations, the same principle applies to tools used for transcription, translation, customer enquiries and production administration. A useful assessment should concern the task being performed, the information involved and the consequences of error. A general statement that a product is “AI-ready” tells a producer little about those details.

### A wider push towards practical governance

A separate UNESCO update published on 2 November describes training for 17 public-sector participants from 16 African and small-island countries alongside its September AI ethics forum. Activities included ethical-impact assessment and tools concerning AI’s environmental effects.

That is a different initiative, rather than a partnership with the Eswatini centre. It nevertheless illustrates a shared practical challenge: turning policy language into people, methods and projects that institutions can sustain.

RIM’s guide to useful professional training asks what participants can actually do afterwards. The same question belongs here. Future reporting should examine which services become available, who can access them and what their assessments demonstrate.

The next milestone is therefore delivery of the announced capabilities. Institutions and businesses should judge progress through published methods and usable services, rather than treating the launch as a completed system of AI assurance.

*Illustration: blank notebook, geometric tiles and a magnifying glass; not an operational laboratory or an actual testing result.*

Why this matters

African organisations need the ability to assess systems in the conditions where they will be used. Building that expertise can strengthen procurement and accountability, including in creative businesses using automated language and administrative tools. Eswatini’s phased approach offers a concrete programme to scrutinise. Its value will depend on competent practitioners and credible evidence; neither training nor the announcement of a centre, by itself, proves that a product is safe or certified.

Sources

Topics: Eswatini · Artificial intelligence · Standards · Professional training · African technology

← All news & insights