AI Governance and Scientific Capability Through the Lens of Evidence

The Democratic Republic of the Congo distributes responsibility among states, companies and scientific actors. It supports cooperation while limiting the Council's technical-regulatory remit. Participation includes resource processing, industrialisation and a place in innovation value chains, not simply technology access. Its security concerns include drones, surveillance and conflict-related value chains. These arguments identify responsibilities and capabilities without independently establishing a governance mechanism's effectiveness.

AfiaGap within the seven-country AI4Mpox collaboration.

AfiaGap within the seven-country AI4Mpox collaboration · Infectious diseases

Active research programme

Implementation reporting does not establish causal health benefit, isolate AI contribution, or establish model performance or prospective external validation; the programme is multinational.

Denmark connects international governance and human responsibility with science-informed policy. It supports substantial UN-centred cooperation and draws attention to digital divides. Military risks remain linked to responsibility, while sovereignty is not materially developed as a separate theme. Its call for evidence identifies a basis for decisions without resolving which evidence supports each decision. Computing infrastructure illustrates enabling capacity, not a demonstrated research outcome.

Gefion AI supercomputer.

Gefion is owned and operated by the Danish Centre for AI Innovation; the Novo Nordisk Foundation holds 85% and the Export and Investment Fund of Denmark holds 15%. The foundation is distinct from Novo Nordisk pharmaceutical company. · Scientific infrastructure

Infrastructure

Gefion is enabling infrastructure; announced research objectives do not establish completed experiments, medicines discovered using the system or patient benefit.

France links human oversight to independent evaluation throughout a system's life cycle, transparency and company liability. It supports substantial UN-centred cooperation and shared evaluation capacity. Its concerns include risks arising from system behaviour as well as deliberate misuse. This approach places responsibility in institutions and companies alongside people. The scientific example illustrates a bounded analytical method; it does not establish that the proposed governance arrangements work.

Deep-learning classification of antibiotic modes of action from bacterial images.

Antibiotic image-analysis researchers; ownership details not supplied · Antimicrobial methods

Computational/methods research

Laboratory-grounded image classification does not establish clinical utility or patient benefit.

Greece argues that AI should assist accountable human judgement, with transparency and cooperation within the Council's security remit. It supports geographically representative and gender-balanced engagement and refers to computing infrastructure and domestic institutions. Compressed decision times are presented as a potential source of miscalculation. An ambiguous automatic-transcript sentence concerning nuclear decisions is excluded from quotation use. The intervention does not independently validate a control mechanism.

BioASQ 2025 biomedical information benchmark.

International BioASQ collaboration · Biomedical information

Computational/methods research

The selected evidence is an international benchmark with Greek institutional participation. It evaluates bounded research tasks, not clinical safety, clinical benefit or Greece’s military-AI safeguards.

Latvia emphasises human direction, control and independent evaluation through capable institutions. Its concerns include deliberate misuse and risks originating in system behaviour. It proposes an informal Council expert group and draws attention to digital divides. Assessment capacity is part of its capability argument. The expert-group proposal remains distinct from Liberia's proposed working group; neither is presented here as an agreed Council mechanism.

AIDA-associated gastric tissue-image methods.

University of Latvia-associated work using pathologist-annotated gastric tissue images, with collaboration involving the Institute of Electronics and Computer Science. The work is associated with AIDA, a multinational EU project coordinated by INCLIVA in Spain; it is not wholly Latvian. · Biomedical imaging

Computational/methods research

Early result is limited to a simple task; no prospective external clinical validation or patient benefit is established. AIDA is multinational, and the topology extension is proposed work.

Liberia calls for independently verifiable safeguards that preserve human agency and support substantial UN-centred cooperation. Participation means enabling Africa to influence standards and architectures, not only to receive technology. It links military risks to responsibility and proposes an informal Council working group. That proposal is separate from Latvia's expert-group proposal. The selected health example supplies observational context, not evidence validating the diplomatic position.

PIH Liberia AI-assisted TB screening pilot and before-and-after report.

Reported implementation involving Partners In Health Liberia and Qure.ai, including local health-service activity and conference-level evaluation evidence. This is mixed and multinational implementation, not exclusive national model ownership. · Public health

Observational/public-health research

Implementation and preliminary before-and-after evidence cannot isolate AI’s contribution; full methods and results are unavailable, and independently validated clinical effectiveness or causal patient benefit is not established.

Pakistan combines human judgement with universally representative governance and substantial UN-centred cooperation. It warns that compressed decision time may increase miscalculation and that the pace of governance must not become technological gatekeeping. Its participation argument resists exclusion. The selected randomised evaluation illustrates the value of testing a specific proposition, including a non-significant primary finding; it does not demonstrate the effectiveness of Pakistan's governance proposals.

SPOT-TB randomised screening-location evaluation; primary benefit not established.

SPOT-TB evaluation team; ownership details not supplied · Infectious diseases

Trial/evaluation

The primary intention-to-treat analysis did not establish improved case-finding yield. More favourable per-protocol findings do not replace it. Results remain a preprint in this evidence base.

Panama stresses that decision-makers should understand AI recommendations, supporting information and uncertainty. It calls for impartial scientific advice, incident exchanges and Global South participation in development and benefits. The intervention also references a national strategy and non-state missile-guidance concerns. These remain attributed arguments. The review has not established a strong selected project-level scientific example, which is a selection gap rather than evidence of national inactivity.

Planned AI/biotechnology conference; no project-level result established.

Panama hosts and participates in a regional framework involving SENACYT, INDICASAT-AIP and the Organization of American States; the biotechnology-centre designation is not itself AI research. · Scientific capacity-building

No strong example established

The reviewed material supports regional biotechnology capacity-building and a planned AI knowledge-exchange event, but no project-level AI-science result; this bounded evidence gap does not establish national inactivity.

The Russian Federation favours specialised, inclusive, consensus-based intergovernmental forums, including established discussions on information security and autonomous weapons, while questioning the generic Council venue. Practical human-control measures are to be determined nationally, with operating constraints, intervention and deactivation. Speed is also presented as a potential advantage. This differs from warnings about compressed decisions elsewhere without implying rejection of safeguards or of international engagement.

ITMO antibiotic–nanoparticle modelling study.

The 2025 study was conducted by ITMO University researchers. The reviewed sources do not establish state ownership or a government-run programme. The separate 2026 ITMO–AIRI joint virtual laboratory is an opened collaboration; it is not the origin of the 2025 study. · Antimicrobial modelling

Computational/methods research

Completed computational modelling only; further experimental verification was future work. No animal or human efficacy, approved medicine or clinically validated antimicrobial treatment is established.