AI enhances scientific decision-making

The Arctic is undergoing dramatic environmental change, with glaciers melting at a rapid rate. The Greenland Ice Sheet’s outflow has increased from about 25 billion tons annually in the 1990s to an average of 234 billion tons today (NASA, 2019). Accelerated glacial melting raises global sea levels, threatening coastal areas and stressing the Arctic ecosystem, and could also disrupt ocean currents, contributing to atmospheric temperature anomalies and affecting the global climate system (Singh, Reference Singh2025). Timely access to data on Arctic glacial melting is therefore important.

Historically, the monitoring of Arctic glacial change relied on manual field surveys, single-point weather stations, and limited satellite remote-sensing data, each of which has notable limitations (NOAA Arctic Program, 2025; World Glacier Monitoring Service [WGMS], n.d.; Zemp et al., Reference Zemp, Frey, Gärtner-Roer, Nussbaumer, Hoelzle, Paul and Vincent2015). Manual observations are constrained by the harsh Arctic environment and the vast territory involved, resulting in limited coverage and high costs (Lee et al., Reference Lee, DeGrandpre, Guthrie, Hill, Kwok, Morison and Wilkinson2022; Starkweather et al., Reference Starkweather, Larsen, Kruemmel, Eicken, Arthurs, Bradley and Petäjä2021). Single-point weather-station data may lack spatial representativeness (Pielke et al., Reference Pielke, Stohlgren, Parton, Doesken, Moeny, Schell and Redmond2000) and could fail to capture overall glacial change (Machguth, Purves, Oerlemans, Hoelzle, & Paul, Reference Machguth, Purves, Oerlemans, Hoelzle and Paul2008). While satellite remote sensing offers macro-scale advantages, manual processing of the resulting data is time-consuming, and prediction accuracy has been limited (Chuvieco, Reference Chuvieco2020). Faced with the Arctic’s complex and variable climate, conventional methods could struggle to capture real-time, precise glacial-ablation dynamics, which may leave policymakers without accurate, time-sensitive data and hinder effective scientific decision-making.

AI offers a potential solution to this dilemma. With its computing capabilities and advanced algorithms, AI could substantially improve Arctic glacier monitoring and prediction (Maslov, Persello, Schellenberger, & Stein, Reference Maslov, Persello, Schellenberger and Stein2025; Shionalyn et al., Reference Shionalyn, Catania, Trugman, Shahin, Stearns and Felikson2025). On one hand, it enables rapid analysis of satellite imagery, helping to capture glacial morphological changes and to convert subtle displacements and thickness variations of ice into data streams that could support a dynamic monitoring network covering the Arctic ice sheet (Hemanth, Reference Hemanth2020). On the other hand, AI could be used to mine patterns from large, long-term climate datasets (for example, temperature and precipitation) to help project future Arctic glacial-ablation trends and improve understanding of Arctic ecosystem dynamics (AMAP, 2021; Post et al., Reference Post, Alley, Christensen, Macias-Fauria, Forbes, Gooseff and Wang2019; Reichstein et al., Reference Reichstein, Camps-Valls, Stevens, Jung, Denzler and Carvalhais2019). AI-generated high-precision Arctic sea-ice monitoring and prediction results could provide reliable data support for scientific decision-making in climate governance (Mehryar, Yazdanpanah, & Tong Reference Mehryar, Yazdanpanah and Tong2024; Rolnick et al., Reference Rolnick, Donti, Kaack, Kochanski, Lacoste, Sankaran and Bengio2022), shipping safety (Wagner et al., Reference Wagner, Hughes, Bourbonnais, Stroeve, Rabenstein, Bhatt and Fleming2020), and other fields (Andersson et al., Reference Andersson, Hosking, Pérez-Ortiz, Paige, Elliott, Russell and Shuckburgh2021; Li, Hsu, & Tedesco Reference Li, Hsu and Tedesco2024).

For coastal nations threatened by sea-level rise associated with Arctic glacial melt, AI-based approaches could help predict sea-level-rise trends and create more detailed climate-risk and coastal flood-risk maps, thereby supporting government planning for flood-control projects, coastal defences, urban drainage networks, and emergency-response strategies (Kulp & Strauss, Reference Kulp and Strauss2018; NASA Sea Level Change Team, n.d.; Zitouni, Elneel, Albakri, Alkhatib, & Al-Ahmad, Reference Zitouni, Elneel, Albakri, Alkhatib and Al-Ahmad2025). AI-derived cryospheric and climate data may also support the transformation of national emission-reduction strategies from broad to more targeted approaches: by integrating glacial-ablation indicators, climate-model simulations, and AI-assisted emissions-monitoring data, governments could better identify high-emitting sectors or sources and formulate more targeted environmental policies to address greenhouse-gas emissions that contribute to Arctic glacial melting (Kaack et al., Reference Kaack, Donti, Strubell, Kamiya, Creutzig and Rolnick2022; Rolnick et al., Reference Rolnick, Donti, Kaack, Kochanski, Lacoste, Sankaran and Bengio2022; Rouet-Leduc, Kerdreux, Tuel, & Hulbert, Reference Rouet-Leduc, Kerdreux, Tuel and Hulbert2024). In addition, AI-based data-fusion systems have the potential to provide earlier warnings based on Arctic sea-ice predictions (Li et al., Reference Li, Hsu and Tedesco2024), which could help Arctic shipping enterprises plan safer routes, avoid ice hazards, and improve transport efficiency while maintaining safety.

Optimising resource exploration for sustainable development

The Arctic is resource-rich yet extremely fragile and vulnerable to even minor disturbances. Global warming has accelerated Arctic glacial melting and placed additional stress on the ecosystem, and poorly regulated resource development has further increased ecological pressure (Canuel, Reference Canuel2016). Constrained by the Arctic’s harsh environment, conventional resource-exploration methods could be inefficient, often requiring large-scale surveys that are time- and labour-intensive and that may disturb permafrost, vegetation, and species habitats. Balancing Arctic resource development with ecological protection has long been a difficult challenge.

AI offers a potential way to address this tension. In pre-exploration planning, AI could rapidly process large geological and remote-sensing datasets to help identify prospective resource-rich areas, improving the precision of exploration and reducing unnecessary fieldwork, resource waste, and potential environmental disturbance associated with poorly targeted exploration (Mern, Corso, Burch, House, & Caers Reference Mern, Corso, Burch, House and Caers2024; Shirmard, Farahbakhsh, Müller, & Chandra Reference Shirmard, Farahbakhsh, Müller and Chandra2022). For example, KoBold Metals has used AI to analyse geological surveys, satellite imagery, and historical data and is reporting having identified the high-grade Mingomba copper deposit in Zambia (Mern et al., Reference Mern, Corso, Burch, House and Caers2024). As the technology matures, AI has the potential to further improve Arctic exploration efficiency and reduce ecological disruption.

When identifying resource zones, AI could be combined with ecological-sensitivity assessment models (Silvestro, Goria, Sterner, & Antonelli, Reference Silvestro, Goria, Sterner and Antonelli2022) to help avoid biodiversity hotspots. By integrating data on Arctic vegetation distribution and animal habitats, such tools could generate ecological-sensitivity maps that guide exploration routes away from high-sensitivity areas and help minimise disturbance to biodiversity.

AI could also play a key role in resource-development processes. Using IoT sensors and drones, it could support real-time monitoring of development areas (Ateya, Algarni, Koucheryavy, & Soliman, Reference Ateya, Algarni, Koucheryavy and Soliman2022): sensors collect soil, water-quality, and atmospheric data, while drones capture high-resolution images. AI could analyse these data to help detect potential ecological problems (for example, vegetation damage, pollution, or oil leaks) and to generate timely alerts and remediation suggestions, helping to minimise harm and support more sustainable resource development.

From pre-exploration through to development, AI could support efforts to balance Arctic resource development with ecological protection, offering an innovative approach to combining efficient development with longer-term ecological sustainability through intelligent tools applied across the full process.

Ushering in a new era of international cooperation

As a region where global climate change is strongly amplified – warming at more than twice the global average (AMAP, 2021) – and as a geopolitical focal point, the Arctic faces significant challenges. Intertwined issues of accelerated glacial melting, resource disputes, and ecological protection have placed considerable strain on existing, fragmented governance arrangements – an assembly of overlapping national jurisdictions and sectoral agreements among the Arctic states. AI technology could help drive a shift in Arctic governance from fragmented, poorly coordinated systems toward data-enabled and participatory co-governance, where shared observing infrastructures and AI-supported decision systems strengthen collaboration among Arctic stakeholders (Kavanagh, Reference Kavanagh2026a; Starkweather et al., Reference Starkweather, Larsen, Kruemmel, Eicken, Arthurs, Bradley and Petäjä2021).

Fragmented and siloed data governance has been identified as a challenge for collective action (Benfeldt, Persson, & Madsen Reference Benfeldt, Persson and Madsen2020), and in the Arctic context, the dispersal of data across national systems could hinder cooperation. For example, Arctic data are often distributed across nationally or institutionally maintained repositories, such as Russia’s permafrost monitoring data, Norway’s marine environmental and fisheries datasets, and Canada’s ice-service records. These repositories vary in data structures, standards, access protocols, and update frequencies, creating the kind of fragmented web of repositories (Demidov et al., Reference Demidov, Anisimov, Anisimov, Borisik, Golavskii, Gusakova and Zaitseva2025; Environment and Climate Change Canada, 2025; Institute of Marine Research, 2025; Yarmey & Khalsa, Reference Yarmey and Khalsa2014). AI-based tools could help integrate such datasets through automated data cleansing, standardisation, and cross-system integration (Reichstein et al., Reference Reichstein, Camps-Valls, Stevens, Jung, Denzler and Carvalhais2019). By combining transnational information – for example, satellite imagery, marine-sensor data, and scientific-expedition records – AI could help reduce information barriers (Ghamisi et al., Reference Ghamisi, Rasti, Yokoya, Wang, Höfle, Bruzzone and Benediktsson2019) and contribute to a more unified, dynamically linked Arctic data resource and a more standardised data system for shared use (Wilkinson et al., Reference Wilkinson, Dumontier, Aalbersberg, Appleton, Axton, Baak and Mons2016).

Beyond data integration, AI could also influence how Arctic states cooperate, potentially supporting a shift from largely independent national efforts toward more collaborative governance. In the past, national assessments may have diverged because of differences in available data and models, which could deepen divisions in cooperation. Where AI systems are designed to be transparent and explainable, they could allow participating states to jointly trace parameters, verify data sources, and check that conclusions could withstand multi-party scrutiny (Rudin, Reference Rudin2019). Such an approach – emphasising visible data, traceable processes, and verifiable conclusions – could help build trust by reducing suspicion arising from information asymmetry and supporting a shift in Arctic governance toward more evidence-based consensus-building. It should be noted, however, that transparency and explainability are not inherent properties of all AI systems and remain active areas of research (Barredo Arrieta et al., Reference Barredo Arrieta, Díaz-Rodríguez, Del Ser, Bennetot, Tabik, Barbado and Herrera2020).

AI-driven real-time early-warning networks could also support transnational emergency coordination, which is important given the Arctic’s ecological fragility. One emerging example is the COSRVA project associated with the Arctic Council’s Emergency Prevention, Preparedness and Response (EPPR) Working Group. This platform is intended to support real-time oil-spill monitoring – detecting leaks and generating cross-border alerts through data processing – and to help evaluate the relative merits of different cleanup strategies (Cook, Reference Cook2025). It is designed to facilitate the sharing of emergency resources and experience and to improve international oil-spill response. It should be emphasised, however, that such systems are still emerging and are not yet widely implemented across the Arctic. Where they are available, such risk-sharing and rapid-response mechanisms could help Arctic states reach actionable consensus based on shared, AI-supported evidence, even where cooperation has previously been constrained by sovereignty disputes.

More broadly, AI could become a useful tool for Arctic cooperation by helping to integrate information, supporting shared analytical approaches, and contributing to the early identification of transboundary ecological risks (Reichstein et al., Reference Reichstein, Camps-Valls, Stevens, Jung, Denzler and Carvalhais2019). By foregrounding shared data and evidence, AI could help mitigate some disputes and foster collective action. Amid escalating climate threats, AI could enhance ecological monitoring and, where used transparently, support more cooperative approaches to international governance. On a question as consequential as the future of the Arctic, shared governance will be essential to sustaining the region, and AI-supported co-governance may form one part of future polar cooperation.

Algorithmic bias and decision deviation

As AI is integrated more deeply into Arctic governance, algorithmic bias and decision deviation could increasingly affect the stability and fairness of international cooperation. The role of AI in Arctic ecological monitoring, resource allocation, and waterway management depends on data collection and model training, and inherent technical limitations raise important issues.

Arctic governance AI systems could be affected by limitations in data collection and model construction. First, the uneven geographical distribution of training data could contribute to imbalances in decision-making (Shankar et al., Reference Shankar, Halpern, Breck, Atwood, Wilson and Sculley2017). Because of harsh conditions and high costs, technologically and financially well-resourced actors tend to dominate monitoring networks, while some Arctic Indigenous communities have more limited data-collection capacity. For example, the Arctic observation network has documented incomplete and uneven spatial coverage (National Research Council, 2006). Such imbalances may bias AI sea-ice prediction models toward better-monitored regions, which could lead to waterway-navigation and expedition-route decisions that do not adequately reflect the needs of less-monitored areas (Gray et al., Reference Gray, Boss, Prochaska, Kerner, Begouen Demeaux and Lehahn2024).

Second, the constraints and effects of data annotation could shape AI outputs (Girard-Chanudet, Reference Girard-Chanudet2023), and subjective or inconsistent labelling may introduce bias. In Arctic ecological protection, AI models often require labelling of species distributions and pollution zones, but cognitive differences among institutions and experts could lead to inconsistent results. For instance, in monitoring methane leakage from Arctic tundra, an over-reliance on satellite remote sensing relative to ground sampling could lead AI models to underestimate local methane risks, which may skew resource allocation and weaken protection for critical ecological areas (Hashemi et al., Reference Hashemi, Räsänen, Virtanen, Juutinen, Grosse, Aurela and Treat2025).

Algorithmic bias could also contribute to unequal resource allocation (Obermeyer, Powers, Vogeli, & Mullainathan Reference Obermeyer, Powers, Vogeli and Mullainathan2019). The traditional ecological knowledge and the needs of Arctic Indigenous peoples may be overlooked by AI systems (Carroll et al., Reference Carroll, Garba, Figueroa-Rodríguez, Holbrook, Lovett, Materechera and Hudson2020). Where training data do not include Indigenous information, such as reindeer migration routes or areas used for fishing and hunting (Kløcker Larsen et al., Reference Larsen, Boström, District, District, District and Wik-Karlsson2022; Persson, Harnesk, & Islar Reference Persson, Harnesk and Islar2017), AI-informed resource-development plans could encroach on the environments these communities depend on and affect cultural heritage. For example, an AI-informed plan for Arctic mineral extraction that did not account for Indigenous land claims could damage reindeer habitat and contribute to conflict between communities and developers (Larsen et al., Reference Larsen, Boström, District, District, District and Wik-Karlsson2022; Persson et al., Reference Persson, Harnesk and Islar2017).

The black-box nature of AI and governance effectiveness

Dependence on advanced technology could bring risks that are difficult to anticipate. In the context of Arctic governance, the wider use of AI raises what is often called the black-box problem. In this commentary, a black box refers to an AI system whose internal decision-making process is not readily interpretable by the people who use it or are affected by it, so that the reasons for a given output are difficult to examine. While AI could improve the efficiency of certain decisions through its data-processing capacity, the limited interpretability of some systems could reduce the transparency of Arctic governance and raise questions about fairness and accountability.

The limited interpretability of many machine-learning algorithms arises from their complex internal structure; reducing this opacity through interpretability and explainability is an active area of research (Mukhamediev et al., Reference Mukhamediev, Popova, Kuchin, Zaitseva, Kalimoldayev, Symagulov and Yelis2022). For example, deep-learning systems use multi-layer neural networks that learn from large datasets and could help address complex problems (LeCun, Bengio, & Hinton Reference LeCun, Bengio and Hinton2015), but their internal workings could be difficult for people to follow (Lipton, Reference Lipton2018). Consider a hypothetical AI system used to support Arctic fisheries management that integrates ocean temperatures, data on fish migration, and historical catch data to generate allocation options, with very large numbers of parameter adjustments during training. Even where the output appears reasonable, the underlying data weightings and decision rules may remain difficult to interpret so that even developers may be unable to fully explain how a particular quota was derived – making it hard for stakeholders to assess fairness.

The limited interpretability of AI could be particularly significant in sensitive areas of Arctic resource allocation. Decisions about oil, gas, and minerals involve multiple stakeholders, but AI systems may give more weight to quantifiable indicators (such as reserves, costs, and returns) than to factors that are harder to quantify (such as Indigenous land rights and traditional ways of life). For example, an AI-recommended large-scale Arctic mineral-extraction plan based primarily on geological and economic data might not adequately account for Inuit ancestral lands or reindeer migration routes. Where the decision process is difficult to interpret, affected Indigenous communities may find it hard to understand or take part in it, which could contribute to a lack of consensus and to delays or conflict.

Environmental protection could also be affected by limited interpretability. The fragile Arctic ecosystem calls for careful decision-making, but the complexity of AI models could limit both the traceability of a decision (the ability to reconstruct how it was reached) and its explainability (the ability to understand why) in ecological-restoration plans (De Bruijn, Warnier, & Janssen Reference De Bruijn, Warnier and Janssen2022). In the governance of Arctic glacial melting, an AI system might analyse data on warming and glacier movement to propose intervention options, but the relevant climate model parameters and the logic linking interventions to predicted effects may be encapsulated within the algorithm (Reichstein et al., Reference Reichstein, Camps-Valls, Stevens, Jung, Denzler and Carvalhais2019). As a result, environmental organisations and governments may be unable to trace how a recommendation was produced or to understand its scientific basis well enough to assess the associated risks (Mittelstadt, Allo, Taddeo, Wachter, & Floridi Reference Mittelstadt, Allo, Taddeo, Wachter and Floridi2016), which could weaken trust, hinder cooperation, and slow ecological-protection efforts.

The use of AI in Arctic governance may also reinforce existing technological asymmetries – a dynamic sometimes described as technological hegemony, in which a small number of actors with advanced AI and data capabilities end up exerting disproportionate influence – which could compound the interpretability problem. Patent analyses indicate that advanced AI capabilities are concentrated among a relatively small number of countries and firms (Wicaksono & Hendrix, Reference Wicaksono and Hendrix2023). In the Arctic context, actors with such capabilities and data resources could come to dominate algorithm-based decision-making, with their systems and models functioning, in effect, as default bases for decisions and potentially marginalising those with fewer technical resources. There is also a risk that, in areas such as Arctic passage rule-making, states with leading AI capabilities – particularly the United States, given its leading position in AI model development and its stated Arctic interests in protecting navigation freedoms and defending US interests in the region – could use their advantages to shape rules in ways that favour their own interests (Maslej et al., Reference Maslej, Fattorini, Perrault, Gil, Parli, Kariuki and Oak2025; The White House, 2022). Where such processes are not transparent, other Arctic states may find it difficult to participate meaningfully, which could undermine the perceived fairness and credibility of governance.

Ethical dilemmas and social acceptance

AI is being integrated into Arctic governance rapidly, creating new possibilities for ecological monitoring and resource development but also raising ethical questions and challenges for social acceptance – that is, the willingness of communities and the public to accept AI-based governance (Robles & Mallinson, Reference Robles and Mallinson2025). Arctic Indigenous peoples, including the Inuit and the Sámi, have lived in the Arctic for generations and hold distinctive knowledge and ways of life (Nuttall, Reference Nuttall2005). The large-scale application of AI could place these communities at the centre of difficult trade-offs between technological development and cultural continuity (Kavanagh, Reference Kavanagh2026b).

AI could improve the efficiency of resource exploration and development, but a technology-driven approach may prioritise economic returns and give insufficient weight to Indigenous peoples’ relationships with the land (Córdova-Pintado, Reference Córdova-Pintado2026; Kuokkanen, Reference Kuokkanen2019). In addition, limited participation by Indigenous peoples in decision-making about how AI is developed and governed could heighten concerns about social acceptance. At present, the development and governance of AI for the Arctic are often led by governments, research institutions, and enterprises, and Indigenous voices may be under-represented in some of these processes – although Indigenous organisations such as the Inuit Circumpolar Council, the Sámi Parliaments, and the Global Indigenous Data Alliance are increasingly active in data governance and related decision-making (Carroll et al., Reference Carroll, Garba, Figueroa-Rodríguez, Holbrook, Lovett, Materechera and Hudson2020; Inuit Circumpolar Council, 2022). Indigenous ecological knowledge – including detailed understanding of seasonal patterns and wildlife behaviour – is not always fully incorporated into the design of AI systems. For example, some Arctic AI ecological-monitoring systems rely primarily on satellite remote sensing and sensor networks and may not fully draw on Indigenous observations of ecological change, although there are also initiatives that incorporate Indigenous knowledge and support Indigenous-led data work (Johnson et al., Reference Johnson, Alessa, Behe, Danielsen, Gearheard, Gofman-Wallingford and Svoboda2015). Where Indigenous peoples are excluded from benefit-sharing, this could contribute to dissatisfaction and to resistance to AI applications, which may affect the social acceptance of governance measures (Robles & Mallinson, Reference Robles and Mallinson2025; Wilson, Reference Wilson2019).

The balance between efficiency and fairness is an increasingly prominent issue in the use of these technologies. Algorithms are typically optimised for data-driven efficiency and for what they treat as optimal resource allocation, but this efficiency-oriented logic could be in tension with social fairness in Arctic governance – for example, with the equitable distribution of benefits and burdens among Arctic states and communities. In allocating access to resources such as fisheries, minerals, and shipping routes, AI systems may favour options expected to yield the greatest economic returns – often to developers and resource-rich actors – while giving less weight to the development needs and cultural rights of Indigenous communities (Loewen-Colón, Reference Loewen-Colón2026). AI may also intersect with existing inequalities: actors with advanced AI capabilities could exert considerable influence over Arctic governance, potentially concentrating that influence (Ishkhanyan, Reference Ishkhanyan2026). Communities and actors with more limited technical capacity or representation – in particular some Indigenous communities, as well as some non-Arctic stakeholders – may find it harder to benefit from technological progress and could be more exposed to negative impacts, which may widen existing gaps (Signé, Reference Signé2026).

If these issues are not addressed, they could hinder the sustainable development of Arctic governance, particularly cooperative, multilateral governance involving both states and Indigenous peoples. Arctic Indigenous peoples are key stakeholders in the region – as stewards of the environment, as rights holders, and, in many cases, as active participants in and beneficiaries of resource development – so their participation and support matter for the legitimacy and sustainability of governance, including both ecological protection and economic development (Barnes, Reference Barnes2026). Continued neglect of their rights and interests could contribute to social conflict and weaken the foundations of international Arctic cooperation, making it harder to achieve shared goals for ecological protection and resource development.

Enhancing the transparency and fairness of AI

As AI is increasingly integrated into Arctic governance, the limited interpretability of black-box algorithms could become a significant barrier to public trust (Rudin, Reference Rudin2019; von Eschenbach, Reference von Eschenbach2021). Across applications such as shipping-route planning, resource-development assessment, and ecological-protection decisions, AI may offer efficient governance tools, but opaque internal logic could raise public concerns about the soundness and fairness of decisions, with the potential to erode trust (de Fine Licht & de Fine Licht, Reference de Fine Licht and de Fine Licht2020; Kizilcec, Reference Kizilcec2016). Addressing this concern will depend in part on promoting transparent algorithm design and on systematically strengthening public confidence in AI-supported governance.

One element of transparent algorithm design is the careful documentation and disclosure of the decision process (Chaudhary, Reference Chaudhary2024). Taking Arctic fishery-resource management as an example, the integration of multiple data sources (such as satellite remote sensing, ocean-monitoring stations, and fishers’ reports), the preprocessing steps (such as outlier removal, format standardisation, and normalisation), and the model operations (such as population prediction or quota optimisation) could be disclosed, including model architecture, parameter choices, and iterative steps. Presenting this information in an accessible, visualised form could help bridge the gap between technical systems and public understanding. It is important to note, however, that the value and limits of algorithmic transparency are debated. While some argue that disclosing and disassembling the decision process could build trust (Chaudhary, Reference Chaudhary2024), others caution that efforts to explain complex models could be misleading and that explainability does not by itself guarantee genuine understanding or accountability (De Bruijn et al., Reference De Bruijn, Warnier and Janssen2022). The distinction between traceability – reconstructing how a decision was reached – and explainability – understanding why – is useful here: disclosure may improve traceability without necessarily making a model’s reasoning interpretable.

Explainable AI (XAI) methods could help surface the basis of algorithmic decisions and support efforts to improve fairness, although explainability alone does not guarantee fair outcomes (Minh, Wang, Li, & Nguyen Reference Minh, Wang, Li and Nguyen2022). In the context of Arctic resource allocation, XAI could help trace the decision logic and identify potential sources of data bias – for example, an over-reliance on monitoring data from a few countries (Taylor et al., Reference Taylor, Lawler, Aronsson, Barry, Bjorkman, Christensen and Schmidt2020) or insufficient incorporation of Indigenous ecological knowledge (Armitage, Berkes, Dale, Kocho-Schellenberg, & Patton, Reference Armitage, Berkes, Dale, Kocho-Schellenberg and Patton2011). Establishing a data-review committee could enable multi-dimensional assessment of data sources and decision rules, helping to ensure that decisions respect Indigenous rights and balance the interests of different states. Embedding principles such as fairness and ecological protection as explicit constraints in data processing, model training, and output could help safeguard fairness in resource allocation. Clearer decision logic and more visible processes could, in turn, support trust and encourage broader participation in governance.

Establishing technical standards and ethical guidelines for AI governance

Existing legal frameworks, which were largely designed for established policy areas such as shipping, fisheries, and environmental protection, were not developed with the full life cycle of AI in mind, which could make it harder to respond to emerging issues such as algorithmic opacity, data-security risks, and ethical conflicts. Encouraging the Arctic Council to develop technical standards and ethical guidelines for AI governance could therefore be valuable for supporting sustainable development and for balancing the interests of multiple stakeholders.

First, technical standards could be developed that integrate technical feasibility with institutional adaptability and that are tailored to the Arctic’s distinctive environment and governance needs. Harsh conditions (such as extreme cold and strong winds) call for clear indicators of equipment environmental adaptability (such as operating-temperature range and protection rating) to support stable long-term operation. For fragmented, multi-source, heterogeneous data (across meteorology, oceanography, and ecology), unified data-format standards – covering collection parameters, storage formats, and transmission-security protocols – could help establish a more standardised process and enable smoother cross-regional and cross-domain data sharing and analysis. Such standards would also need to be compatible with existing Arctic governance systems to avoid implementation obstacles.

Second, ethical guidelines could be developed that centre on sustainable development and respect for Indigenous rights. AI applications should prioritise ecological protection to avoid damaging the fragile Arctic ecosystem – for example, by using ecological-carrying-capacity assessments as core indicators in resource development decisions in order to balance economic growth with ecological protection. Guidelines should explicitly require respect for Indigenous cultural traditions, knowledge systems, and rights, support the full participation of Indigenous peoples in data collection and decision-making, and promote equitable sharing of the benefits arising from AI. They should also address AI-related risks (such as privacy and algorithmic bias) so that technology is aligned with fairness, justice, and transparency.

Multi-stakeholder participation is important for ensuring that any standards and guidelines are authoritative and enforceable. As the central Arctic governance platform, the Arctic Council could take a leading role, bringing together governments, research institutions, enterprises, and Indigenous organisations. In such a process, governments might set out regulatory requirements, research institutions provide technical input, enterprises offer feasibility feedback, and Indigenous organisations represent community interests. A dynamic mechanism for monitoring and evaluation could help identify and address problems as they arise and support effective implementation, providing a stronger institutional basis for Arctic AI governance. Together, such measures could help regulate AI applications, mitigate risks, build consensus, and support a more sustainable pattern of Arctic governance.