Every day, thousands of images and signals collected at sea – sonar, buoys, satellites, cameras installed on ships – generate enormous amounts of data. Artificial intelligence (AI) is already being used to interpret such data from, for example, detecting dolphins in real time to estimating biodiversity indicators, or automatically identifying species caught onboard fishing vessels to improve fisheries management models or resolve disagreements in marine spatial management. But behind this technological transformation emerges a key question: can we fully trust what AI says when the health of the ocean is at stake?
A European team led by AZTI – Marine and Food Research (Spain) has developed a framework that establishes three pillars for marine AI to be reliable, ethical, and scientifically robust. While AI adoption is accelerating worldwide, global AI governance in the marine domain remains fragmented, with differing regulatory approaches across regions. The full work is detailed in the scientific article “Towards Trustworthy Artificial Intelligence for Marine Research, Fisheries and Environmental Management”, was published in Fish and Fisheries.
“We are seeing a massive increase in the use of AI algorithms that process the vast streams of marine data – from cameras and sonar to satellite observations – but they often fail to meet expectations,” explains José A. Fernandes, AZTI AI expert and lead author of the study. “The key question is: how much trust can we place in the AI algorithms? Given that AI is already a reality for the fishing and marine research sector, it will only be useful if it is trustworthy. Our work establishes how to ensure trustworthiness by combining science, ethics, and industry engagement.”
A real problem: when algorithms fail
AI offers enormous possibilities, but also risks. An onboard camera system used for automated catch monitoring, for example, can mistake two similar species if it hasn’t been trained by experts and with images taken under diverse lighting conditions. A model that predicts fish abundance can fail if it is built on incomplete or biased data, giving a misleading picture of the real state of a population. Automated tools may also face resistance within the industry if their decision-making processes aren’t transparent or fail to reflect the practical knowledge of those who work at sea. These examples illustrate why robust criteria for quality, transparency, and validation are essential, especially in a field where decisions affect ecosystems, fishing communities, and public policy.
Three pillars for an AI that builds trust
The framework proposed by the research team is structured around three main pillars. The first focuses on socioeconomic and legal viability. The development and use of AI must be accessible to the entire marine sector, including small-scale fisheries, and be aligned with international and regional regulations-such as the European Union’s new AI Regulation – to ensure global coherence and fairness in implementation. The study emphasizes that the most effective tools are those designed with the direct participation of stakeholders, and not solely for them, which increases social acceptance, incorporates local knowledge, and reduces resistance.
The second pillar concerns the ethical governance of data. For AI to function effectively, it needs diverse, clean, traceable, and responsibly managed datasets. The authors recommend applying FAIR, CARE, and TRUST principles to marine data, ensuring that information-images, sensor signals, or monitoring records – is interoperable, respectful of the communities generating it, and preserved for long-term use. Good data governance, they argue, is the foundation for transparency, reproducibility, and accountability.
“When AI is used to guide decisions that affect marine ecosystems and livelihoods, accessibility, transparency and validation are essential,” says Catarina Silva, co-author and researcher at the University of Coimbra, Portugal. “Our framework provides practical guidance to ensure that AI strengthens scientific evidence and trust across the marine sector.”
The third pillar addresses technical robustness and scientific validation. AI must demonstrate its reliability under real-world ocean conditions – not just in controlled environments. The study recommends validating models with independent data, applying statistical tests, and comparing outcomes with on-site measurements. For instance, automated catch
analyses can be checked against manual port sampling to identify discrepancies. Such cross-validation ensures that algorithms reflect reality and deliver genuinely useful management tools.
Benefits for research, fishing, and society
The framework’s implications extend to the scientific community, administrations, the fishing sector, and the public.
For marine research, it provides coherent criteria for developing and benchmarking AI models, improving comparability, and accelerating insights into ecosystem health and climate impacts. For fisheries and environmental management, it strengthens the reliability of decision-support systems-from quota allocation and marine spatial planning to the monitoring of illegal fishing. Properly validated models and well-governed data can help optimise routes, reduce emissions, enhance traceability, and improve sustainability at sea. For society, trustworthy AI ensures that ocean digitalisation proceeds responsibly. It supports a sustainable blue economy, balancing technological innovation with social and ecological well-being. As AI becomes increasingly integrated into environmental governance, the authors stress that regulation and ethics must evolve alongside technology.
“This means that the use of AI in catch monitoring technologies for fisheries must also actively contribute to support the implementation of global commitments such as the Kunming – Montreal Global Biodiversity Framework, the BBNJ Agreement, and the provisions of the emerging plastics treaty. Integrated AI solutions on fishing vessels and across marine monitoring systems can only contribute to these goals if they are developed within a co-management framework – bringing scientists, policymakers, industry, and fishing communities together to ensure transparency, accountability, sustainability, and even practical implementation onboard vessels. By engaging industry directly in data collection and validation, AI becomes not merely a tool for surveillance, but a governance instrument that translates practical, equitable data collection from the ocean into concrete action that is aligned with international biodiversity and sustainability targets, even possibly enabling more adaptive and evidence-based area-based management and protection over time.”
Rachel Haug Fossbakk (formerly Tiller), Chief Scientist
Director of SINTEFs strategic research area of Biodiversity and Area Use
Department of Fisheries and New Biomarine Industry
SINTEF Ocean
The study (https://doi.org/10.1111/faf.70052) was published in the journal Fish and Fisheries, one of the world’s leading journals in marine and fisheries science, which publishes interdisciplinary research at the interface of ecology, management, and policy.