Responsible AI Governance: Innovation, Ethics and Sovereignty — A Contribution to FGI-CMR 2026

Written contribution to Panel 3 of the Cameroon Internet Governance Forum (FGI-CMR 2026) — Palais des Congrès, Yaoundé, 18 – 20 August 2026.

FGI-CMR 2026, under the High Patronage of the Prime Minister, Head of Government
Central theme: “From fragmentation to interoperability: making data the foundation of digital public infrastructure in Cameroon”
Panel 3: Responsible governance of artificial intelligence: innovation, ethics and sovereignty

Mr Armand Gaetan NGUETI, expert panellist — Founder & CEO, UBTS International Corp (United States) · T20 Delegate — G20 Summit (South Africa, 2025) · Africa President, Government Blockchain Association · AfriDES Executive Coordinator for Central Africa · Chairman of the Board, PIPRA Cameroun S.A.

1. Executive summary

Cameroon does not, first and foremost, have an artificial intelligence problem: it has a trusted-data problem. No sovereign AI system can be built on data that is fragmented, untraced and unauditable. Responsible AI governance therefore begins upstream of the algorithm — at the source of the data, in its provenance, its quality and its auditability.

This contribution puts forward six key ideas, aligned with the thematic axes of Panel 3 and with the expected outcomes of its work:

Trusted data comes before the model. Traceability must be designed in from the outset (traceability by design), not reconstructed after the fact. It is the first condition for the quality of training data and for tackling bias.
Endogenous AI is a doctrine of operational sovereignty, resting on three pillars: data sovereignty, vernacular AI — that is, the integration of national languages — and the strategic resilience of infrastructure.
Auditability is organised through a maturity framework, an instrument already proven in the governance of distributed ledger technologies and transferable to the assessment of AI systems used by public administrations.
Compliance is a strategic asset, not a constraint. The full entry into application of the law on the protection of personal data, on 23 June 2026, gives Cameroon a legal foundation that can be used immediately to frame AI — provided the two workstreams are explicitly linked.
Priority must go to applied AI and agentic AI, anchored in the industrial sanctuaries and structuring pillars of the Industrialisation Master Plan, rather than to generative AI. The current over-focus on the latter diverts attention from productive uses and turns an import-substitution policy into a policy of importing digital services.
International cooperation must be organised on non-extractive terms. What circulates globally should be norms, standards and engineering; what remains national is the data, its governance and the value it produces. Fair sharing with local stakeholders is not an ethical extra: it determines the technical reliability of the data collected.

These ideas lead to seventeen operational recommendations, set out in section 10 and addressed by name to the stakeholders concerned.

2. The contributor’s standing and the basis of this contribution

This contribution draws on a threefold experience — in the field, in global governance and in teaching — whose nature should be made clear, because it determines the scope and the limits of the positions put forward.

Operationally, the author is the founder and chief executive of UBTS International Corp, a Delaware C Corporation based in the United States that builds digital infrastructure and trusted data for Africa, and which designed the pan-African UBTS Africa Agritech Digital Passport (UAA-DP) programme and its national version, the Cameroon Agritech Digital Passport (CA-DP), whose launch in Cameroon will follow the Ivorian launch, in smart agriculture and its value chains. He is also the inventor of CAMTRADE PASS, a socio-technological initiative operated by PIPRA Cameroun S.A., whose Board of Directors he chairs: a traceability solution combining blockchain, artificial intelligence and the internet of things, serving the “Made in Cameroon” label, the fight against counterfeiting and compliance with GS1 standards. The pilot was carried out with a Cameroonian manufacturer, integrating fifty-six product references and a “one product, one QR code” labelling scheme enabling real-time tracking, authenticity checks and engagement with the end consumer.

In international governance, the author has chaired since December 2024 the global Food Supply Chain Working Group of the Government Blockchain Association — the first African appointed to this role — where he leads the work on the “food chain” supplement to the Blockchain Maturity Model, in collaboration with the Dynamic Coalition on Blockchain Assurance and Standardization (DC-BAS) of the United Nations Internet Governance Forum. He is, to date, the first and only certified BMM consultant from the African continent. As a T20 delegate to the G20 Summit held in South Africa in 2025, he contributed public-policy expertise to the Think20 engagement group; he is also executive coordinator of the African Digital Economy Summit for Central Africa.

Finally, in teaching, a Microsoft Certified Trainer since the age of twenty-four and president of the international association of Microsoft Certified Trainers for Africa, the author recently designed and delivered, at the BGFI Business School in Libreville, the first Master’s-level (Master 2) course on digital assets and central bank digital currencies ever taught in Central Africa. This experience bears directly on the panel’s “research, innovation and skills” axis.

The purpose of this section is not to showcase a career, but to establish the evidential basis for the statements that follow: each recommendation below stems either from an observed deployment or from a governance mandate actually held.

3. The diagnosis: data fragmentation comes before the algorithmic question

The central theme chosen by the Forum — moving from fragmentation to interoperability and making data the foundation of digital public infrastructure — rightly identifies the starting point. All its consequences for artificial intelligence should be drawn.

3.1. An inverted value chain

Public debate on AI naturally focuses on models, algorithms and their performance. Yet in the Cameroonian context, as in most African economies, the limiting factor is not the model: it lies upstream, in the availability of structured, lawful, representative and verifiable data. An AI system trained on data whose provenance cannot be established produces results whose reliability cannot be established either. Data quality is therefore not a technical prerequisite: it is the governance question itself.

3.2. Three concrete forms of fragmentation

Institutional fragmentation: public data remains spread across administrations with no common reference for identification, format or quality, which rules out any national-scale model training and reproduces the silos at the algorithmic level.
Documentary fragmentation: in economic value chains, particularly agri-food chains, information on a product’s origin, composition and journey is still recorded on heterogeneous media, often paper, not time-stamped and not legally enforceable — and therefore unusable by an automated system.
Linguistic fragmentation: the available text data overwhelmingly reflects French and English, whereas a substantial share of economic and social interaction takes place in national languages. Any model trained on this incomplete corpus mechanically reproduces exclusion.

3.3. The central risk: sovereignty proclaimed without substance

The result is a risk that must be named plainly: digital sovereignty asserted in texts but lacking any material substance. An administration can host its processing on national soil and still depend on models trained elsewhere, on foreign data, carrying representations that are not its own. Locating servers in the country is a necessary condition; it is never a sufficient one. Sovereignty is decided by control of the corpus as much as by control of the infrastructure.

4. Two national use cases: CAMTRADE PASS and the Cameroon Agritech Digital Passport

The panel’s “use cases” segment calls for concrete, national demonstrations. Two are presented here, complementary in their position along the value chain: the first concerns the industrial and commercial downstream, the second the agricultural upstream. Their interest for this discussion lies not in the technology used but in the method they validate.

4.1. CAMTRADE PASS: what the system establishes

The principle is simple: each product receives a unique digital identity, embodied in a code the consumer can read, and backed by a tamper-proof record of its journey. Blockchain for integrity and time-stamping, the internet of things for capture at source, artificial intelligence for flow analysis and anomaly detection: the combination produces data that is not merely collected but qualified — that is, data whose origin, date and integrity can be verified by a third party.

4.2. Three lessons transferable to public action

Traceability must be native. It is far less costly to record provenance when data is created than to reconstruct it later. The principle applies equally to data produced by administrations: civil registers, land data, health data, customs data.
Standardisation is the vehicle of interoperability. Basing the system on GS1 standards is not a technical detail: it is what makes Cameroonian data readable by the information systems of its trading partners. Data that is sovereign but unreadable outside has no exchange value.
Trust is built through verifiability, not declaration. The consumer, importer or regulator who can verify for themselves does not need to believe. This shift — from declarative trust to verifiable trust — is at the heart of what responsible governance of automated systems should be.

4.3. The Cameroon Agritech Digital Passport: carrying trusted data all the way to the agricultural source

The pan-African UBTS Africa Agritech Digital Passport programme, designed by UBTS International Corp, follows the same principle but applies it where it is hardest to establish and most useful: in the agricultural upstream. Its national version, the Cameroon Agritech Digital Passport, will launch in Cameroon after the Ivorian launch, in smart agriculture and its value chains. The principle is to give agricultural products a digital identity attached from the farm, carrying the origin, the cropping practices and the product’s journey through to the processor or exporter.

Three reasons make this positioning strategically decisive for the panel’s work.

It tackles fragmentation where it is greatest. As noted in section 3.2, information on origin and farming practices is still overwhelmingly recorded on heterogeneous media, not time-stamped and not enforceable. No agricultural data policy is possible until this information is structured at source.
It is the substrate without which AI applied to agriculture remains theoretical. Yield forecasting, early detection of crop diseases, demand anticipation, optimisation of industrial supply: none of these uses, discussed in section 5.2, is achievable without a structured, dated and reliable database on national production. The digital passport is not an artificial intelligence project: it is the condition that makes one possible.
It sits exactly where agro-pastoral import substitution is measured, whose acceleration is among the recommendations of the mid-term review of the National Development Strategy. A value chain whose volumes, qualities and calendars are documented becomes one from which a manufacturer can source locally with the predictability it previously got from imports.

Finally, the programme’s architecture deserves attention in its own right: a pan-African framework ensuring consistency of standards and interoperability at continental scale, broken down into national sub-programmes that keep control of their data and adapt to their value chains. It is the exact operational translation of the Forum’s central theme: moving beyond fragmentation without giving ground on sovereignty, with interoperability built through shared standards rather than the centralisation of data.

4.4. Transfer to priority sectors

The same architecture can be transferred, without any conceptual break, to several of the sectors identified as priorities: traceability of the medicines chain and the fight against counterfeit pharmaceuticals; certification of study paths and diplomas in the education system; traceability of agricultural inputs and production, extending the logic of the agritech digital passport; and securing public procurement through tamper-proof time-stamping of documents. In each case, AI does not come first: it comes once the data has been made reliable.

5. Priority to applied and agentic AI: anchoring artificial intelligence in structural transformation

The panel is to give its view on priority sectors for the use of artificial intelligence. This contribution argues that prioritisation is not first a question of sectors but of technology families: which artificial intelligence do we fund, and to produce what?

5.1. An over-focus that shifts the debate

Since 2023, public debate on artificial intelligence — in Cameroon as elsewhere — has narrowed very largely to a single branch: generative AI. Chat assistants, text and image generation, deepfakes: these take up most of the media, institutional and budgetary attention. This focus is not illegitimate in itself — vigilance about deepfakes and disinformation is fully justified, and generative AI holds real potential for processing national languages. It nonetheless produces three effects that must be named.

It establishes the idea that artificial intelligence is a communication and content technology, whereas in the economies that draw measurable value from it, it is first and foremost a production technology.
It steers spending towards consuming services designed and hosted abroad. A country that measured the success of its AI policy by the adoption rate of foreign chat assistants would in fact have deepened its digital-services deficit. There is a paradox here that should be stated clearly in an economy that has made import substitution the central lever of its development strategy: financing imported artificial intelligence in the name of modernisation simply swaps one dependency for another.
It leaves outside public policy the uses that create local added value, reduce identified import lines and rely on data the country controls.

5.2. What applied AI covers

Applied AI refers to systems trained on the data of a specific production process to solve a defined problem whose value can be measured. It is less spectacular than generative AI and considerably more profitable. Its uses are documented and transferable: quality control by machine vision at the end of the line; predictive maintenance of industrial equipment and energy infrastructure; demand forecasting and inventory optimisation; detection of counterfeits and anomalies in trade flows; agricultural yield forecasting and early detection of crop diseases; load optimisation on electricity grids; targeting of customs and tax inspections through risk analysis.

These uses share one decisive feature: they are trained on data produced in Cameroon, on Cameroonian processes, and their results convert directly into competitiveness or reduced imports. In other words, they are the exact technological translation of the first pillar of the National Development Strategy 2020–2030.

5.3. Agentic AI: the lever for organisations with limited staff

Agentic AI refers to systems that do not merely produce content but carry out chains of tasks under supervision: they plan a sequence, act through software tools, check the result and report back. For administrations and small and medium-sized enterprises structurally short of qualified staff, this is probably the largest productivity reserve of the decade: preliminary processing of customs clearance files, completeness checks on authorisation requests, documentary compliance checks, monitoring of suppliers and markets, reconciliation of data between non-interoperable information systems.

This family, however, calls for a higher standard of governance, not a lower one — and this is where this contribution intends to be rigorous. The risk associated with generative AI is erroneous content; the risk of agentic AI is an erroneous act. A system that acts engages the responsibility of the administration that uses it. Three safeguards must therefore be put in place at the same time as it is deployed: complete, time-stamped logging of the actions taken; technical reversibility of those actions; and mandatory human validation for any act that is irreversible or enforceable against a third party. Promoting agentic AI without these three conditions would be reckless; putting them in place, on the other hand, allows it to be deployed without delay.

5.4. Anchoring in the Industrialisation Master Plan

The link with national industrial doctrine already exists and does not need to be invented. The Industrialisation Master Plan structures the rebuilding of Cameroonian industry around three national industrial sanctuaries — digital, agro-industry and energy —, five structuring industrial pillars — textiles-clothing-leather, mining-metallurgy-steel, forestry-wood, hydrocarbons-petrochemicals-refining, chemicals-pharmaceuticals —, two foundations of emergence, namely infrastructure and financing, and finally a cross-cutting component dedicated to strategic intelligence and economic intelligence.

This last component deserves particular attention: strategic intelligence and economic intelligence are, by their very definition, applied artificial intelligence functions. The Industrialisation Master Plan therefore made room for AI in the national industrial architecture long before the debate on generative AI emerged. That room remains to be occupied.

The mid-term review of the National Development Strategy, published by the ministry in charge of the Economy, concludes that implementation has been resilient but below initial ambitions, and recommends in particular accelerating agro-pastoral import substitution and strengthening technical and scientific training. Applied AI is not an add-on to this programme: it is a direct accelerator, provided it is directed at the value chains concerned rather than at content uses. The selection, in consultation with the private sector, of a first group of companies set to become national champions — chosen on criteria of industrialisation, import substitution, value creation, competitiveness and innovation, mainly in agro-industry — offers an immediately available testing ground. These companies have an industrial process, hence production data; they have a quantified import-substitution target, hence an indicator; they receive public support, hence a framework. They meet the three conditions for a national applied-AI programme with measurable results.

5.5. An operational decision criterion

To keep prioritisation from remaining merely declarative, it is proposed to adopt a simple criterion, applicable to any artificial intelligence project seeking public funding. Four questions, to which the project owner must answer in writing:

Which identified import line does this project help reduce, and by what measurement indicator?
What local added value does it create, and in which segment of the value chain?
What data does it rely on, and does Cameroon control its production and storage?
What national skill does the project leave behind once completed?

A project unable to answer these four questions is not necessarily without merit; it simply does not fall within the strategic priority. The criterion has the advantage of making prioritisation verifiable, and therefore auditable — which is consistent with the overall spirit of this contribution.

6. The principles: the doctrine of endogenous AI

The National Artificial Intelligence Strategy, presented in July 2025 by the Ministry of Posts and Telecommunications, sets an explicitly sovereigntist ambition and includes, among its published orientations, the development of multilingual models integrating national languages. This orientation aligns very directly with the doctrine the author champions within the international alliance for sovereign AI under the name of endogenous AI. It is proposed to set out its three pillars, so that they can inform the national governance principles expected from this work.

6.1. First pillar — data sovereignty

It is defined less by where data is stored than by control of the chain: knowing what data is produced on the territory, by whom, under what legal regime, for what purposes it is reused and on what terms it may be transferred. Cameroon’s legal framework provides decisive support here: the law on the protection of personal data makes transfers outside the territory subject to prior authorisation by the Protection Authority. This mechanism, designed to protect privacy, is also — and this reading should be embraced — an instrument of industrial data policy.

6.2. Second pillar — vernacular AI

An artificial intelligence system that does not understand the language in which a citizen makes a request does not serve that citizen: it sorts them. Linguistic inclusion is therefore not a nice-to-have but a requirement of equality before public services. It calls for a documented effort to build corpora in national languages, with a clear legal status — the national language corpus should be treated as a digital public good, built with public funding and with organised access, rather than as a resource captured by whoever collected it first.

6.3. Third pillar — strategic resilience

Resilience means guaranteeing the continuity of digital public services regardless of decisions taken by outside actors: diversified suppliers, contractual reversibility written in from the call for tenders, the ability to run at the network edge when connectivity is degraded, and documentation of critical dependencies. An administration that does not know what would break if a given supplier stopped its service is not exercising governance: it is exercising hope.

7. Innovation, ethics and sovereignty: a model of global co-creation and fair orchestration

The title of this panel brings together three terms that current debate readily sets against each other: innovation is supposed to require openness, ethics constraints, sovereignty closure. This contribution argues that, properly linked, they are three faces of one organisational model — and that the experience of the programmes mentioned above makes it possible to define its terms.

7.1. The false dilemma between sovereignty and cooperation

No country in the world, however powerful, builds its entire artificial intelligence value chain alone. Digital sovereignty understood as technological autarky is not an ambitious project: it is an unachievable one, and its predictable failure serves as an argument for those who advocate total dependency. The relevant question is therefore not whether to cooperate, but on what terms.

There is indeed a mode of cooperation that weakens sovereignty while presenting itself as a partnership. Its pattern is constant: an outside entity collects local data, processes it and trains its models outside the territory, then sells the resulting service back to the actors who supplied the raw informational material. Local stakeholders appear in it as suppliers of raw data, never as co-owners of the value created. The pattern is all the harder to challenge because it comes with real financial transfers and virtuous communication. It must nonetheless be called what it is: extractive cooperation.

7.2. What the UBTS model puts to the test

The experience led by UBTS International Corp is of interest well beyond this particular case. A Delaware C Corporation based in the United States, with access to global standards, funding and standard-setting networks, designs a pan-African programme — the UBTS Africa Agritech Digital Passport — whose implementation is entrusted to national sub-programmes, such as the Cameroon Agritech Digital Passport, rooted in the value chains and institutions of the country concerned.

This configuration reverses the extractive pattern on one decisive point: what circulates globally is norms, interoperability standards and engineering; what remains national is the data, its governance and the value it produces. Cooperation concerns the framework, not the resource. That is exactly the dividing line any public policy on artificial intelligence partnerships should adopt.

7.3. Four conditions for non-extractive cooperation

It is proposed that the panel adopt four cumulative conditions, which could be written as a standard clause into any public partnership involving data or artificial intelligence systems:

Data ownership and jurisdiction. Data produced on national territory remains the property of the actors who generate it — producers, cooperatives, companies, administrations — and falls under Cameroonian jurisdiction, any transfer being subject to the prior authorisation regime established by law.
Documented skills transfer. The partnership includes a training component with verifiable results: number of people trained, skills acquired, ability to operate the system independently at the end of the cooperation. A partnership that leaves no skills behind is not cooperation: it is a service contract.
Sharing of the value created. Upstream actors — first among them agricultural producers and their organisations — capture an identifiable share of the value that traceability makes possible: access to new markets, quality differentiation, reduced losses, easier access to finance.
Reversibility and no lock-in. The use of open, documented standards ensures that the system will outlive the partnership that created it and can be taken over, audited or replaced without being rebuilt from scratch.

7.4. Social responsibility as a technical condition, not an add-on

Corporate social responsibility is too often relegated to institutional communication. In agricultural data systems, on the contrary, it is a condition for the system to work, and this point deserves to be put before the panel because it directly connects the ethics axis with the data-quality axis.

A digital agricultural passport system only works if producers feed it data. And producers only feed it data over time if they draw a tangible benefit from it. Fair value sharing is therefore not a moral requirement added on top of the technical system: it is what determines its reliability. An unfair system produces incomplete, late or false data — and, for that reason, becomes technically deficient.

This proposition has general reach. Wherever data is produced by dispersed and poorly equipped actors — agriculture, crafts, local commerce, community health —, the quality of the training data for future artificial intelligence systems will depend on the fairness of the architecture that collects it. The linguistic inclusion discussed in section 6.2 follows the same logic: a system that addresses producers only in a language they do not master captures impoverished information. Here, ethics is not the price paid for performance; it is its precondition.

7.5. Fair orchestration of stakeholders

That leaves the question of orchestration: who convenes, arbitrates and sustains over time a system bringing together producers, cooperatives, manufacturers, regulators, research institutions and technical and financial partners? The State alone would have to commit considerable administrative resources; an outside platform would impose its own terms. Experience suggests a third way: a dedicated orchestration structure, with an explicit social-responsibility identity, whose role is to hold the table rather than take the best seat at it.

This model is not foreign to the forum hosting this work — on the contrary, it is its founding principle. Since the World Summit on the Information Society, the Internet Governance Forum has rested on the joint participation of governments, the private sector, the technical community, academia and civil society. Cameroon applies this principle to deliberation; it is proposed to extend it to execution, by making multi-stakeholder governance the operating model of national sectoral data programmes, and not only their mode of discussion.

8. Audit, evaluation and experimentation: proposing a maturity framework

The panel’s expected outcomes include recommendations on mechanisms for auditing and evaluating AI systems. This is where the author believes he can make the most specific contribution, because of his mandate within the Government Blockchain Association.

8.1. The lesson of the maturity model

The experience of the Blockchain Maturity Model shows that an assessment framework is only useful if it has four characteristics: it is graduated — it does not merely set compliant against non-compliant, but grades levels of maturity; it is publicly documented, so that those assessed know the rule before the assessment; it is applied by certified assessors, which requires prior training; and it is revisable, at intervals set in advance. A framework lacking these four conditions produces formal compliance, not quality.

8.2. Proposal: a national maturity framework for public AI systems

It is proposed to develop a national assessment framework applicable to any AI system deployed by an administration, organised around six dimensions: the provenance and quality of training data; the explainability of decisions produced and the ability of the person concerned to contest them; documented measurement of bias, notably linguistic, geographic and gender bias; the security and resilience of the system; the system’s degree of autonomy and the reversibility of the actions it takes — a dimension specifically intended for the agentic systems discussed in section 5.3; and finally a clear chain of responsibility, that is, identification of the natural or legal person answerable for the decision.

8.3. The link with existing law

This framework would not start from a blank page. Law No. 2024/017 of 23 December 2024 on the protection of personal data, whose transitional compliance period ended on 23 June 2026, already requires prior impact assessments for processing likely to present high risks to the rights and freedoms of individuals. Almost every AI system deployed by an administration falls into this category. The data protection impact assessment is therefore the natural legal anchor for evaluating AI systems: rather than creating a parallel mechanism, it is proposed to enrich the existing impact assessment with an AI-specific component. This solution has the double advantage of an immediately available legal basis and a marginal administrative burden.

8.4. The regulatory sandbox

Supervised experimentation — the regulatory sandbox — makes it possible to reconcile innovation and risk control, under three conditions that are rarely all met: a limited and explicitly defined scope; a fixed duration with a mandatory evaluation at the end; and an organised exit, whether generalisation, adjustment or termination. Without an exit clause, a sandbox becomes a permanent exemption; this is the main pitfall observed in comparable schemes.

9. National capabilities: data, infrastructure, skills and cooperation

9.1. The scale of the training effort

The targets announced at the launch of the National Strategy include, by 2040, training several tens of thousands of artificial intelligence specialists, with a stated goal of greater gender parity, and creating several thousand direct jobs. Such an effort cannot rest on initial university training alone: it requires industrialising continuing education, certifying the trainers themselves and recognising professional experience.

The programme run in July 2026 at the BGFI Business School in Libreville, where practising bank executives were trained in digital assets in fifteen hours without writing a single line of code, illustrates a workable path: job-oriented teaching, grounded in regional cases, designed for complete beginners and leading to an ability to make professional judgements. This format can be replicated in Cameroonian administrations at a controlled cost.

9.2. Training assessors, the forgotten condition

One point deserves particular attention because it is regularly overlooked: an audit framework without trained assessors remains a dead letter. Experience with the maturity model in distributed ledger governance shows that certification of assessors must come before, not after, the assessment scheme takes effect. It is therefore recommended to start building a national pool of AI system auditors as early as the framework’s drafting phase.

9.3. International cooperation as a lever, not a dependency

This Forum is the national chapter of a global process stemming from the World Summit on the Information Society. This lineage offers a concrete and under-used opportunity: the standard-setting work carried out within the dynamic coalitions of the United Nations Internet Governance Forum — notably on assurance and standardisation — is a channel through which a Cameroonian contribution can be carried to the global level, not merely received. The author has direct access to it and is at the authorities’ disposal to relay national positions there. Likewise, participation in the G20’s Think20 engagement group offers a channel of influence over international orientations on technology governance.

The shift to make is this: stop importing norms that one applies, and start exporting use cases that set the norm. An operational national traceability system is, in this respect, as much a diplomatic asset as an industrial one.

10. Operational recommendations

The following seventeen recommendations are worded so that they can be taken up directly in the panel’s conclusions. They are listed in order of operational priority.

Enrich the data protection impact assessment, already required since 23 June 2026, with a component specific to artificial intelligence systems, rather than creating a parallel assessment scheme. Lead addressee: MINPOSTEL · Data Protection Authority.
Develop a graduated national maturity framework for public AI systems, built around the six dimensions set out in section 8.2 and published before it takes effect. Lead addressee: MINPOSTEL · National AI governance body.
Build and certify a national pool of AI system auditors, with assessor training preceding the framework’s entry into force. Lead addressee: MINPOSTEL · Higher education institutions.
Write the requirement of native traceability (time-stamped, verifiable provenance) into the specifications of every public project producing data intended for model training. Lead addressee: All administrations · Public procurement.
Recognise the national language corpus as a digital public good, organise its building with public funding and define its access and reuse regime. Lead addressee: MINPOSTEL · MINESUP · MINAC.
Base national data systems on international identification and exchange standards, since external readability determines the exchange value of national data. Lead addressee: MINPOSTEL · MINCOMMERCE · ANOR.
Establish a regulatory sandbox with a defined scope, a fixed duration and a mandatory exit clause evaluated at the end. Lead addressee: MINPOSTEL · ART · Data Protection Authority.
Require, in every public contract for an AI system, a reversibility clause and documentation of the supplier’s critical dependencies. Lead addressee: Public procurement · All administrations.
Prioritise four demonstration sectors with strong ripple effects: pharmaceutical traceability, diploma certification, agri-food chains and securing public procurement. Lead addressee: Prime Minister’s Office · MINPOSTEL · Sector ministries.
Bring a structured Cameroonian contribution to the dynamic coalitions of the United Nations Internet Governance Forum, to promote national use cases at the global standard-setting level. Lead addressee: MINPOSTEL · National representation at the IGF.
Adopt a decision criterion applicable to every AI project seeking public funding: import line reduced, local value added, national control of the data used, skills left behind. Lead addressee: MINEPAT · MINFI · MINPOSTEL.
Explicitly link applied artificial intelligence to the strategic and economic intelligence component of the Industrialisation Master Plan, and to the three national industrial sanctuaries and five structuring pillars. Lead addressee: MINMIDT · MINEPAT · MINPOSTEL.
Launch a national applied-AI programme for the national champions selected under the structural transformation agenda, with priority on agro-industry and a quantified import-substitution target. Lead addressee: MINEPAT · MINMIDT · Private sector.
Specifically regulate agentic AI systems deployed by administrations: time-stamped logging of actions, technical reversibility, and mandatory human validation for any act that is irreversible or enforceable against a third party. Lead addressee: MINPOSTEL · Data Protection Authority.
Include a standard non-extractive cooperation clause in every public partnership involving data or AI systems: national ownership of and jurisdiction over data, documented skills transfer, identifiable sharing of the value created, reversibility through open standards. Lead addressee: MINPOSTEL · MINEPAT · MINREX.
Recognise digital agricultural passport programmes as public data infrastructure for the agro-pastoral sector, a prerequisite for any use of AI in value chains and a means of measuring import substitution. Lead addressee: MINADER · MINEPIA · MINEPAT · MINPOSTEL.
Extend the IGF’s multi-stakeholder model from deliberation to execution, by entrusting the orchestration of national sectoral data programmes to structures bringing together producers, manufacturers, regulators, research and technical partners. Lead addressee: MINPOSTEL · Sector ministries · Private sector.

11. Conclusion

Cameroon comes to this Forum in a singularly favourable position, which should be appreciated. Since July 2025 it has had a national artificial intelligence strategy explicitly oriented towards sovereignty and linguistic inclusion. Since 23 June 2026 it has had a fully applicable data protection framework, with a supervisory authority and instruments — impact assessments, prior authorisation of transfers — that can be used directly to frame AI. And it has national use cases whose logic of verifiable traceability can be extended to priority sectors.

What is missing is therefore neither vision, nor law, nor technical demonstration: it is an explicit link between these three assets. This panel is precisely the place where that link can be made. The most economical and most immediately actionable recommendation of this contribution fits in one sentence: do not build an AI assessment regime alongside data protection law, but build it as an extension of that law.

A second link, just as decisive, remains to be made: between artificial intelligence policy and industrial policy. As long as AI is thought of separately from structural transformation, it will remain an expense; anchored in the industrial sanctuaries, the structuring pillars and the import-substitution objective, it becomes an investment whose return can be measured. The decisive choice of the coming years will not be about how open to be to artificial intelligence, but about the kind the country funds: AI that produces content, or AI that produces value.

Finally, there is the question of approach. Cameroon does not have to choose between opening up and protecting itself: it has to set the terms on which it opens up. The programmes carried out in smart agriculture show that cooperation can be global in its standards and sovereign in its data, provided that fairness towards local stakeholders is treated not as a token clause but as the condition for the system’s own reliability. That, in the final analysis, is the proposal this contribution puts to the Forum: sovereignty is not opposed to cooperation; it sets its terms.

Responsible governance of artificial intelligence is not decreed in a text: it is proven by a citizen’s ability to check for themselves what a system has decided about them, and by a State’s ability to know where the data that informs it comes from.

12. Sources and methodological note

Forum reference documents: letter of invitation No. 0000167/MPT/CAB from the Minister of Posts and Telecommunications; Panel 3 technical brief (general objective, specific objectives, problem statement, thematic axes, guiding questions, format, expected outcomes).
National framework: Law No. 2024/017 of 23 December 2024 on the protection of personal data, promulgated on 23 December 2024, whose eighteen-month transitional compliance period ended on 23 June 2026; National Artificial Intelligence Strategy presented on 7 July 2025 at the Palais des Congrès in Yaoundé during the second national consultations on artificial intelligence.
Strategic and industrial framework: National Development Strategy 2020–2030 (SND30), whose first pillar concerns the structural transformation of the economy and makes import substitution its central lever; mid-term review of the SND30 published by the ministry in charge of the Economy; Industrialisation Master Plan (PDI), structured around three national industrial sanctuaries, five structuring industrial pillars, two foundations of emergence and a strategic and economic intelligence component.

Armand Gaetan NGUETI — Expert panellist, Panel 3, FGI-CMR 2026 — Yaoundé, August 2026.