Eswatini Launches AI Safety And Standards Centre

Eswatini has launched an AI Safety and Standards Centre of Excellence intended to build national expertise in the testing, governance and responsible use of artificial intelligence.

The initiative brings together the Eswatini Standards Authority, known as ESWASA, and AICE Africa. Their partnership was formalised in September 2026, while the programme was publicly announced on 2 November. According to the programme announcement, ESWASA will provide institutional leadership, standards expertise and national coordination. AICE Africa will contribute programme design, governance, platform development and technical skills transfer.

The centre is significant because AI assurance is often discussed after systems have already entered workplaces and public services. Standards bodies can instead ask earlier questions: what evidence supports a model's claimed performance, which failures have been tested, who remains accountable for a decision, and whether a system works for the people and conditions it is meant to serve.

The programme is planned in four stages. The first focuses on ESWASA's internal transformation, including an EswasaOne platform expected to connect standards, certification, inspection workflows and management oversight. A second stage will develop practitioners such as standards officers, auditors, regulators and technical personnel. Industry services and regional cooperation are intended to follow.

Some of the most ambitious elements remain proposals rather than operating facilities. The partners envisage a research and testing laboratory for AI safety, along with future certification services. Those services will require validated testing methods, qualified personnel, institutional impartiality and, where appropriate, accreditation. Announcing a centre does not by itself create the competence to certify complex systems.

That distinction should shape how progress is measured. The first useful evidence will be a published governance structure, a curriculum for practitioners, operational standards projects and documented tests on real systems. A credible centre should also disclose the limits of its assessments. Passing one benchmark cannot guarantee that a model will remain safe after deployment, when data, users or operating conditions change.

Eswatini's standards infrastructure gives the initiative a practical institutional home. ESWASA's mandate already covers standardisation, quality assurance and conformity assessment. Extending that experience into AI could help the country avoid treating responsible technology as a purely voluntary promise made by vendors.

The centre can also serve a regional purpose. Many African regulators face the same difficulty: global AI standards are developing quickly, but local agencies have limited laboratories, computing resources and specialists able to interpret them. Shared methods and training may be more efficient than every country attempting to build a complete assurance system alone.

A regional summit is planned for July 2027, although the dates have not been confirmed. By then, the strongest contribution would be more than a declaration. Eswatini should be able to show which standards it is using, which practitioners have been trained, how tests are governed and what institutions can learn from the results.

The initiative is therefore best understood as the start of a capability-building programme. Its promise lies in moving AI safety from general principle to repeatable institutional practice. Its credibility will depend on whether the proposed laboratory, training and certification functions are built transparently and evaluated against real public needs.

Sources: the programme announcement and ESWASA.