BBNJ Agreement's Clearing–House Mechanism: AI for good?

1 Introduction

The Agreement under the United Nations Convention on Law of the Sea for the Conservation and Sustainable Use of Biological Diversity of Areas beyond National Jurisdiction (BBNJ Agreement or Agreement)¹ entered into force on 17 January 2026.² The Agreement will use conservation and sustainable use measures, such as Area-based management tools, including Marine protected areas (ABMTs, including MPAs),³ and environmental impact assessments (EIAs)⁴ to ensure long-term conservation and sustainable use of nearly two-thirds of the global ocean that is outside of any country's national jurisdiction (Humphries et al., 2025). The Agreement also includes provisions that will enable the fair and equitable sharing of benefits of Marine Genetic Resources (MGRs)⁵ and foster capacity-building and the transfer of marine technology (CB&TMT).⁶The Agreement will also ensure that relevant rights associated with traditional knowledge of Indigenous Peoples, and local communities are protected (Pena-Neira and Coelho, 2025).

Transparency in the implementation of the BBNJ Agreement is paramount, and the principles, institutional arrangements, and processes for enabling such concept are embedded throughout the Agreement (Gottlieb et al., 2025a). In particular, one of the critical institutional arrangements of the Agreement is the Clearing-House Mechanism (Mechanism), mainly an open-access information system that allows users to provide and access information relevant to the Agreement.⁷ The Mechanism will support information sharing associated with the four substantive elements of the Agreement (i.e., MGRs, ABMTs, including MPAs, EIAs, and CB&TMT) as well as the secretariat functions and other matters.⁸ The information system could also bolster funding and finance, and would protect the rights of Indigenous Peoples, and local communities that hold traditional knowledge (Gottlieb et al., 2025b).

The Mechanism will be used to handle and manage information and data ranging from meeting dates to more complex topics, such as those associated with research activities in areas beyond national jurisdiction (ABNJ) (see text footnote 8). The complex web of functions of the Mechanism requires careful contextual analysis on how end-users will access and provide information and data in designing the Mechanism (Gaebel et al., 2025). The options on the design, operation, and maintenance of the Mechanism are under consideration by the Preparatory Commission (PrepCom) at the United Nations (2025) (A/AC.296/2025/19). Once operational, the Mechanism will replace the interim measures implemented by the interim secretariat to exchange relevant information.⁹

The use of advanced IT technologies in the Mechanism has not been confirmed by the PrepCom, but the use of artificial intelligence (AI) is mentioned in the Consolidated draft study on the technical aspects of the operationalization of the Clearing-House Mechanism under the Agreement under the United Nations Convention on the Law of the Sea on the Conservation and Sustainable Use of Marine Biological Diversity of Areas beyond National Jurisdiction (Consolidated Draft Study on the Mechanism), issued for the third PrepCom session (United Nations, 2026). In the Consolidated Draft Study on the Mechanism, AI is mentioned as an overlay to the Mechanism after it is operational, which would have the potential to improve user experience without impacting the underlying authority or core workflows. This article presents that while the use of AI could improve user experience of an information system that has both open and restricted information and data, there are actual, potential and/or perceived risks on the use of AI that will need to be managed with safeguards.

Section 2 discusses foundational information on AI. Section 3 explores some of the key safeguards on the use of AI in an information system. Section 4 draws on insights gained by one of the authors (M.S.M.), as part of his affiliation with the Ocean Matcher Foundation. The section provides an illustrative example of an information system (Ocean Matcher) that uses AI for matching donors and recipients on projects that focus on the ocean, and discusses some of the “lessons learned” that could be informative for the use of AI in the Mechanism. Section 5 provides some of the considerations in incorporating AI into the Mechanism's CB&TMT functions, with a focus on needs assessment stages. In Section 6, this article proposes minimum requirements for the use of AI in reducing actual and potential risks in the Mechanism. In Section 7, two issues are considered. First, some considerations on perceived risks and ways to combat misinformation are discussed. Second, this article elaborates on a potential role for Indigenous Peoples and local communities to lead the efforts to co-develop safeguard processes associated with traditional knowledge or to provide advice on prohibiting the use of AI on and information associated with traditional knowledge. Section 8 provides a summary and a conclusion that the policy question is not whether AI should or should not be used but under what minimum conditions its use is consistent with implementation of the Agreement and can be trusted by the Parties and other key stakeholders, such as Indigenous Peoples and local communities. Further, this paper proposes that the Mechanism has an opportunity to go beyond minimum conditions and lead the way in setting new industry standards for best practices on safeguards on AI use in information systems.

2 What is AI, and what type of technologies may be useful for the BBNJ Agreement's Clearing–House Mechanism?

Artificial intelligence (AI) has advanced rapidly in recent years, driven by the emergence of large language models (LLMs) such as ChatGPT, Gemini, and Claude, which now play a prominent role in public and policy discussions [Stanford Institute for Human-Centered Artificial Intelligence (HAI), 2024]. These models exemplify “foundation models,” large, general-purpose systems trained on vast datasets that support a wide range of downstream applications [Bommasani et al., 2021; Stanford Institute for Human-Centered Artificial Intelligence (HAI), 2024].

AI systems are best understood as computational tools designed to perform tasks associated with human cognitive capabilities, including pattern recognition, prediction, and structured decision-support (Jordan and Mitchell, 2015; Russell and Norvig, 2021). Rather than constituting a single unified technology, AI encompasses diverse algorithms and architectures that identify statistical regularities in data and generate probabilistic outputs [Russell and Norvig, 2021; Stanford Institute for Human-Centered Artificial Intelligence (HAI), 2024].

Modern AI systems do not possess autonomous judgment or moral agency. While there is no universal definition of AI, according to the OECD's widely adopted definition, AI systems produce outputs (e.g., predictions, recommendations, or decisions) based on human-defined objectives, with oversight requirements appropriate to the context of use (OECD, 2019a,b). Academic literature similarly emphasizes that current AI systems lack independent reasoning or contextual understanding; instead, they operate by optimizing learned statistical correlations (Floridi and Cowls, 2019; Russell and Norvig, 2021). In practice, this means that AI may support risk assessment, such as estimating earthquake likelihood using seismic data (Kong et al., 2019), but responsibility for issuing evacuation orders or other public actions remains with human authorities.

Even when AI applications automate routine adjustments, such as smart thermostats or industrial control systems, they function through predefined rules, optimization parameters, or feedback loops rather than self-directed reasoning (Åström and Murray, 2010). Maintaining human-in-the-loop or human-on-the-loop structures is widely regarded as essential to ensuring accountability, safety, and alignment with social and legal norms (Floridi and Cowls, 2019; OECD, 2019a,b; Rahwan et al., 2019). In this sense, AI is best understood as an augmentative technology that enhances human decision-making, for example, by synthesizing large volumes of information and data into more accessible or actionable forms, rather than for AI to generate the underlying information or data.

Within the context of the Mechanism, AI's scalability (i.e., its ability to provide consistent performance while serving increasing numbers of users) could be viewed as a way to improve efficiency and accessibility for all users. Digital governance literature shows that appropriately designed algorithmic systems can expand service reach when combined with oversight and accountability safeguards (Dunleavy et al., 2006; Rahwan et al., 2019). In this article, equity refers to whether all users, including those with limited resources, are able to participate on equivalent footing. In the Mechanism context, equity can be evaluated by whether AI-supported tools reduce administrative burdens, support standardized and comparable submissions, and apply transparent and easy to use search or matching criteria with appropriate human oversight.

3 Safeguards for using AI

If AI-enabled functions are incorporated into the Mechanism, they could be used as decision-support overlays to support users, on top of an information system with certain industry-standard safeguards. Given the Mechanism's diverse user community and the possibility of handling restricted or confidential information, safeguards for use of AI must be designed to reduce harmful bias, protect confidentiality and security, and maintain accessibility for users with varying digital capacity (Binns, 2018; OECD, 2019a,b; Jobin et al., 2019).

Safeguards in using AI are most effective when built into both the technical architecture and the institutional governance of the platform. Such effective design of an IT architecture with AI requires early engagement with users and clear limits on what AI can do without human validation (Rahwan, 2019; OECD, 2019a,b). For the Mechanism, three safeguards are particularly important: human oversight, transparency and traceability of outputs, and robust governance with monitoring and audit processes.

First, human oversight means that AI outputs remain advisory. Human review should be required for any content that could influence official records or resource allocations (e.g., needs assessments, CB&TMT reports, or matching recommendations, etc.), consistent with accountability-focused AI management standards, such as ISO/IEC 42001(2022) and (Rahwan, 2019). Human oversight is also a practical response to mitigate known failure modes of large language model (LLM) tools, including fluent but incorrect outputs (“hallucinations”), performance changes as data and contexts shift, and adversarial manipulation of data or inputs (Ji et al., 2022; National Institute of Standards and Technology (NIST), 2023).

Second, transparency and traceability are necessary so that AI-assisted outputs can be understood, challenged, and corrected. At minimum, the system should disclose when AI is used, identify the sources used to generate an answer (e.g., specific documents or data entries within the Mechanism), and preserve an audit trail that supports later review (Doshi-Velez and Kim, 2017; Jobin et al., 2019; National Institute of Standards and Technology (NIST), 2023). For generative features, linking outputs to underlying sources and enabling verification reduces the risk that users identify an AI response as authoritative when it is uncertain or wrong (National Institute of Standards and Technology (NIST), 2024).

Third, robust governance requires defined roles and responsibilities for data stewardship, security, and quality assurance, together with procedures for access control, incident response, and periodic reviews. These reviews should be capable of detecting emerging bias, security vulnerabilities (including data poisoning), and degradation in performance over time (OECD, 2019a,b; National Institute of Standards and Technology (NIST), 2023). Taken together, these safeguards provide the analytic frame applied in Section 4 to evaluate Ocean Matcher's design choices and their relevance for the Mechanism's CB&TMT's functions.

4 An illustrative example of using AI in an information system: Ocean Matcher

Ocean Matcher is an information system intended to address the highly fragmented nature of funding for projects associated with the ocean. The concept for such a platform originated at the Ocean Decade Foundations Dialogue in 2023¹⁰ where a diverse group of stakeholders, including politicians, venture capitalists, investors, and philanthropists identified the need to improve matching funders to donors in a way that reduces administrative burden for both sides.¹¹ Specifically, the Ocean Matcher partnership found the following gap: funders spend significant time identifying credible initiatives, while recipients spend substantial effort searching for relevant funders. The matchmaking platform uses AI as a support tool to structure information, identify potential connections, and improve the quality and relevance of matching between ocean projects, funders, and ecosystem actors. The platform is not designed to make autonomous funding decisions. Final interpretation, validation, engagement, and any funding decisions remain with the relevant human users, funders, and partners to identify relevant opportunities and reduce the need for repeated outreach across the funding ecosystem.

Ocean Matcher uses AI that enables natural language processing and semantic algorithms to connect project proposals with suitable funders. Unlike some of the traditional open calls which require funders to sift through hundreds of applications, the platform is designed to present curated matches based on clearly defined criteria. Such design is intended to reduce administrative burden for funders and increases visibility for project developers. Streamlining matching provides an illustrative example of using AI with the intent to reduce transaction costs to accelerate the flow of resources to where they are most needed.

To that end, Ocean Matcher uses three ways to streamline the administrative processes:

• Project proponents conduct onboarding process by completing standardized profiles describing missions, objective, and needs from the match. Such process is intended to reduce information asymmetry and ensures comparability as well as limits the risks created by incomplete or inconsistent data. Registration is guided to ensure that there is sufficient support that makes the onboarding process user-friendly.
• The platform applies natural language processing and semantic algorithms rather than relying on simple keyword searches. Such method allows AI to recognize nuance in project descriptions and funding preferences, producing matches that are more precise and context sensitive. It also reduces the risk of reinforcing existing funding patterns, which may bias certain applications, a common limitation of quantitative filters.
• Both funders and project developers retain control over the information and when it is disclosed. Intellectual property and sensitive details remain protected until explicit consent is given. Such process is intended to reduce the risk of misuse and build confidence among participants who might otherwise hesitate to share proposals.

Ensuring trust and accountability are essential in matchmaking and requires that a platform, by design, incorporates fairness, security, and transparency. In this regard, as discussed in Section 3 above, three safeguards are particularly relevant: human oversight, transparency and traceability of outputs, and robust governance with monitoring and audit processes. In the case of Ocean Matcher, the following safeguards are embedded in the design of the platform:

4.1 Human oversight

Ocean Matcher is built around human oversight. AI generated outputs are used to support discovery, analysis, and matching, but they are reviewed and interpreted by people. Project owners, funders, and Ocean Matcher facilitators remain responsible for assessing relevance, context, strategic fit, and follow up. In practice, this means that AI helps surface potential opportunities, but it does not replace human judgement. The platform is intended to strengthen informed decision making, not automate decisions that require contextual, scientific, financial, or strategic assessment.

4.2 Transparency and traceability of outputs

Ocean Matcher aims to make outputs transparent by showing users why a project, funder, theme, geography, or opportunity may be relevant. The matching logic is based on structured information such as project descriptions, thematic priorities, geographies, funding needs, organization type, and stated interests. Where AI is used to classify, structure, summarize, or match information, the underlying data points remain traceable to enable users to understand the basis for the output. Such design is important both for trust and for quality control, because users are able to understand not only what is suggested, but also why it is suggested.

4.3 Governance, monitoring, and audit processes

Ocean Matcher is designed with governance and accountability as core principles. The platform is intended to operate as a neutral enabling infrastructure for the ocean ecosystem, and not as a black box decision system. Key safeguards include:

• Human review of important outputs and matches before they are used in convenings, funder dialogues, or strategic recommendations.
• Ongoing monitoring of output quality, including whether matches are relevant, useful, and aligned with user expectations.
• Continuous improvement of the matching logic based on user feedback, expert input, and observed outcomes.
• Internal review processes for data quality, categorization, and potential bias in how projects and funders are represented.
• Clear separation between AI supported recommendations and actual funding or partnership decisions, which remain with the relevant human actors.

As the platform develops further, Ocean Matcher intends to strengthen formal governance around AI use, including documented processes for monitoring, auditability, data quality, user feedback, and responsible use of generated outputs. Continuous review and updating the technical and processes are critical aspects of ensuring that a platform stays up to date with industry best practices and continue to achieve the mission of the platform.

Ocean Matcher is still in its early phase, and the most recent monitoring data from the platform administrator shows that its database includes 833 ocean related initiatives and 47 funders, including philanthropic foundations and impact investors. Through the platform's matching process, more than 35 connections have been established that have progressed into active funding or collaboration discussions.

5 Considerations in incorporating AI into the mechanism's functions associated with the BBNJ Agreement's CB&TMT requirements

The BBNJ Agreement has a general framework for CB&TMT,¹² but more granular elements that would be pertinent for operationalizing the provisions are needed and require careful considerations (Harden-Davies et al., 2024). For example, the Agreement mandates that CB&TMT must “be based on and be responsible to the needs and priorities of developing States Parties, taking into account the special circumstances of small island developing States and of least developed countries, identified through needs assessments on an individual case-by-base, subregional or regional basis” (see text footnote 12) but it does not provide processes on how to conduct a needs assessment. Rather than to view such absence as a missed opportunity, we argue that the text of the needs assessment requirement in the Agreement provides room for flexibility, which is important given the unique circumstances of each Party that conducts needs assessments.

While not tailored to the BBNJ Agreement's CB&TMT nor to a particular region or a State, IUCN provides an illustrative example on the key stages of needs assessment (Stephen and Triraganon, 2009), which is adapted in Table 1 below. As summarized, the stages include numerous steps and stakeholders, analysis, assessment, matching, and documentation of action plan, strategy, and review and reflection. The process could be arduous, especially if all the steps need to be completed manually.

Table 1. Illustrative example of the key stages of needs assessment.

In terms of conducting a needs assessment, the BBNJ Agreement states that “needs and priorities may be self-assessed or facilitated through the capacity-building and transfer of marine technology committee and the Clearing-House Mechanism.” (see text footnote 12) Inclusion of the Mechanism in the context of the CB&TMT creates space for using AI-assisted tools that could support structured needs assessments stages as discussed in Table 1, while preserving the ability of a Party to manage information control with institutional oversight by the secretariat.

5.1 An illustrative example of a needs assessment process supported by AI

AI functions could be designed strictly as a support function and not replace Agreement-mandated review and recommendation making process by the CB&TMT committee or the decision making process by the Conference of the Parties.

An illustrative example of a needs assessment procedural sequence could be as follows:

1. A Party completes the on-boarding process.

2. A Party enters its needs and priorities using a standardized template within the Mechanism, aligned with the needs assessment stages (i.e., planning for the process, situation analysis, action analysis and capacity assessment) identified in Table 1, and uploads relevant supporting documentation. As discussed above, a Party or a group of Parties could upload a needs assessment without using a standardized template to meet their unique situation.

3. An AI-assisted tool, with information provided by the Party and data contained within the Mechanism, as appropriate, generates a draft summary, identifies incomplete fields, and suggests standardized categories or tags to improve comparability to facilitate information that could be used by the CB&TMT committee to recommend potential matching to the Conference of the Parties.

4. The Party reviews, edits, and approves the submission; only the Party may authorize the final version and determine applicable confidentiality or access settings.

5. Following submission, the secretariat conducts quality and completeness checks, ensures appropriate classification and access controls, and maintains a version-controlled record of submissions and revisions. Any AI-generated outputs remain advisory and AI has no independent decision-making authority.

6. The CB&TMT Committee could review individual or aggregated assessments and issue recommendations on matching in accordance with its mandate to the Conference of the Parties.

7. Should there be a match, the recipient Party proceeds to complete an action plan, strategy, review and reflection (or their unique method) to complete the needs assessment stages. All documents are submitted to the Mechanism.

8. Parties may update their assessments, action plan, strategy and review and reflection (or information based on their unique method), and the Mechanism records revisions, subject to the safeguards described in Section 3.

5.2 Considerations on reducing the CB&TMT administrative burden

For CB&TMT recipients and donors, the three processes that Ocean Matcher uses, as discussed above, namely onboarding guidance, semantic matching, and confidentiality, would clearly benefit Parties from a better user experience and would be invaluable to support the value proposition of the Mechanism. The challenge with CB&TMT is that there has been a strong need that has yet to be fulfilled (Gaebel et al., 2025). Therefore, desirability in using the Mechanism is perhaps the most critical aspect in ensuring the success of CB&TMT.

CB&TMT encompasses a wide range of matters, including skills, knowledge, data, financial, and equipment.¹³ Types of CB&TMT also cover a significant spectrum, from development and strengthening of relevant infrastructure, including equipment to the development and strengthening of institutional capacity and national regulatory frameworks or mechanisms.¹⁴ Appendix II of the Agreement uses an inclusive language (i.e., “may include but are not limited to”), and regarding the “indicative and non-exhaustive list” of the types of CB&TMT, the Conference of the Parties are required to “periodically, as necessary, review, assess and further develop and provide guidance…to reflect technological progress and innovation and to respond and adapt to the evolving needs of States, subregions and regions.”¹⁵ Such flexibility embraced in the Agreement also presents the need for the Mechanism to remain agile.

As discussed in the illustrative example using Ocean Matcher, a structured onboarding may be an important aspect of buy-in from participants. Guided registration to complete standardized profiles would reduce information asymmetry, ensure comparability, and limit the risks created by incomplete or inconsistent information and data. The guide in that regard could be done in a user-friendly manner, so that it is easy to understand and would not cause issues for those located in low-bandwidth locations. Thoughtful structured onboarding may be particularly important because CB&TMT requires cooperation and specific modalities, which include considerations for the special circumstances of Small Island Developing States (SIDS) and of Least Developed Countries (LDCs).¹⁶

Further, to ensure that there is as wide a participation as possible, Parties are required to cooperate “at all levels and in all forms, including through partnerships with and involving all relevant stakeholders, such as, where appropriate, the private sector, civil society and Indigenous Peoples and local communities as holders of traditional knowledge, as well as through strengthening cooperation and coordination between relevant legal instruments and frameworks and relevant global, regional, subregional and sectoral bodies.”¹⁷ Such a mandate that involves numerous stakeholders with diverse interests and needs would greatly benefit from streamlined processes to ensure that participation would not lead to burdensome obligations.

5.3 Considerations for a regional approach to CB&TMT: an illustrative example of the SIDS Knowledge Exchange Network

If successful, a streamlined process using AI may facilitate various partnership and initiatives. As a subproject of Oceans 5 Capacity Building initiative for SIDS, a concept for a platform to provide capacity building for small island developing States to better implement the BBNJ Agreement requirements was proposed (SIDS Knowledge Exchange Network (KEN); Gobin, 2025). The platform is envisaged as a foundational capacity tool: centralized, well-designed and a SIDS-owned knowledge exchange architecture (Gobin, 2025). Specifically, KEN is intended to provide and facilitate access to information, including the following: administrative and institutional frameworks; policy and legal guidance; marine and deep ocean science data; funding and partnership opportunities; best practices; new BBNJ scientific information; workshops; meetings; publications, and graphics. KEN may also be developed to provide peer-to-peer training (e.g., webinars) learning, tools, guidance and access to local, regional and global experts.

Knowledge Exchange Network may have various SIDS users, such as BBNJ government representatives, BBNJ negotiators, regional experts on scientific, legal and/or technical matters, and regional research institutions (Gobin, 2025). As proposed, KEN would be designed to assist SIDS to achieve equitable implementation of the Agreement by addressing the highest priority gaps by calling on its regional experts and institutions and accessing relevant and up-to-date information. KEN is still at a conceptual stage, and appropriate funding would be required to ensure that the needs envisioned by the concept are fulfilled (Gobin, 2025).

Pulling resources from various SIDS users has benefits and challenges. One clear benefit is to foster positive synergies, considering limited resources. Needs can be collectively assessed and met across SIDS without re-inventing the wheel. Further, new ideas could emerge as initiatives and partnership are born. To enable coordination and cooperation among SIDS that desire to work as a group, the Mechanism could be used as a tool to facilitate communication and organization of relevant information to KEN. That said, balancing the competing demands of other developing countries, including the least developed countries with the special needs of SIDS, may lead to discussions at the Conference of the Parties on how and when KEN could be integrated into the Mechanism.

5.4 Considerations on CB&TMT matching under the agreement and the use of AI

In terms of matching, the criteria and the process of allocating CB&TMT to the recipients have not been decided. In theory, there could be different processes for matching. For example, the Mechanism could be designed for direct matching where the donors and partners could be matched by the Mechanism, and the donors to the recipients that need CB&TMT and vice versa. Such direct donor to recipient or recipient to donor process is similar to Ocean Matcher. Another process would involve the CB&TMT committee to obtain information and data on the recipients that need CB&TMT and donors and partners that are interested in providing CB&TMT, then the CB&TMT committee would provide a recommendation to the Conference of the Parties, which would decide on the matching. In either scenario, the use of AI could be helpful to reduce the administrative burden for the donors and partners, recipients, and the CB&TMT committee. As it did for Ocean Matcher, AI would sift through information and data, which could be a tedious process, and present a list of matches to both relevant donors and recipients directly, or to the CB&TMT committee.

Use of AI to support the decision-making process of the donors and recipients or in the case of CB&TMT committee, to come up with a list of recommended matches to the Conference of the Parties in a streamlined way (e.g., comparison of candidates, etc.) and with greater compatibility to their goals in a less resource intense manner would be a step forward in fostering CB&TMT. As incorporated in the design of Ocean Matcher, the use of natural language processing and semantic algorithms would give greater chances to the recipients for their project to be noticed because the method recognizes nuance in project descriptions and funding preferences, producing matches that are more precise and context sensitive. As discussed above in Section 4, another benefit of incorporating AI's semantic matching capabilities is the reduction of bias that could favor certain recipients over others because a well-designed semantic matching would reduce the risk of reinforcing existing funding patterns, which is a common limitation of quantitative filters. Further, ensuring protection of confidential or sensitive information will also boost the confidence of the donors and recipients, increasing their likelihood to participate in CB&TMT.

5.5 Considerations on human oversight

While AI may be useful for automating some of the processes, human oversight would be needed to support making recommendations that can be brought to the Conference of the Parties. For example, in addition to the list of recommended matches, the requirements for the CB&TMT committee to make recommendations for follow-up activities (including on how CB&TMT could be further enhanced), would need to be conducted with careful human oversight. The illustrative example of Ocean Matcher's human oversight processes may be useful in considering the Mechanism's human oversight processes.

One benefit of AI as a supportive tool is that the CB&TMT committee's workload would be lessened because the AI tools could produce the information that the committee could base their recommendations on in a manner that is objective, based on data and information provided by the Parties. AI excels at synthesizing information and data. For that reason, an AI function in the Mechanism could support the following review and monitoring activities that the CB&TMT committee is required to conduct based on the reports submitted by the Parties:

• Assess and review needs and priorities of developing States Parties;
• Review the support required, provided and mobilized, and gaps in meeting assessed needs of developing States Parties; and
• Measure performance based on agreed criteria and review results-based analyses.

Also, the Agreement requires that Parties submit CB&TMT reports to the CB&TMT committee.¹⁸ To that end, AI tools could accelerate report creation by synthesizing various documents related to the Party's initiatives and activities associated with CB&TMT that can be reviewed and edited, as needed, by the responsible individual(s). For both the needs assessments and the CB&TMT reports, AI could be particularly useful, if the Parties decide to produce reports at the regional level because the volume of information would be more, and there would be additional coordination among multiple Parties.

6 Minimum requirements for the use of AI in reducing actual or potential risks in the mechanism

Based on the functions discussed in Section 5 above and safeguard considerations elaborated in Section 3, to allow AI tools to address actual and potential risks, the following minimum requirements in the Mechanism are desirable from a practical and testable perspectives:

• Human approval for consequential outputs: Any AI-generated content used for official reporting, published needs assessments, or matching recommendations should require designated human review and approval prior to release or use in CB&TMT committee's considerations (ISO/IEC 42001, 2022).
• Transparency and traceability: The Mechanism should clearly indicate when and how AI is used and preserve an audit trail of AI-assisted outputs, including the underlying documents or data entries used to generate summaries or recommendations, in order for the outputs to be verified and corrected (OECD, 2019a,b; Doshi-Velez and Kim, 2017).
• Data governance and access controls: The Conference of the Parties could provide clear rules for data classification, access permissions, and confidentiality, particularly where sensitive information may be submitted. Where appropriate, options to exclude sensitive content (including traditional knowledge) from AI-assisted processing should be available.

If the above conditions are met, AI-assisted features may have a role in a more easy-to-use and responsive Mechanism for fostering CB&TMT, while preserving the transparency and accountability expectations that meet the requirements of the Agreement. A successful design and implementation of AI-assisted features will also allow for more complex scenarios, such as regional coordination and cooperation on CB&TMT, which would benefit developing States Parties, especially the SIDS and LDCs to maximize the use of limited resources.

7 Perceived risks and co-development of AI safeguard processes with traditional knowledge holders

Due to the relatively new use of AI in information systems and its rapid evolution, there may be perceived risks of AI tools, mainly due to lack of knowledge on the functions and available safeguards associated with the tools. Should the Conference of the Parties decide to add an AI overlay to the Mechanism to enhance user experience, it will be imperative that the secretariat conduct awareness raising activities to ensure trust and uptake in the use of AI in the Mechanism, especially on the safeguards discussed in Section 3 and specific to the Mechanism in Section 6. Such effort by the secretariat could include in-person or virtual workshops, side events, or information papers so that the users understand the specific safeguards that are built-into the design and implementation of AI in the Mechanism.

As for traditional knowledge, should the Conference of the Parties decide that they would be handled by the Mechanism, one way to ensure trust and accountability would be to co-develop the processes for safeguards with the Indigenous Peoples and local communities. Pacific Small Island Developing States issued a proposal in December 2025 and re-introduced it during the third session of the PrepCom to establish the Advisory Mechanism on Indigenous Peoples, and traditional knowledge (Advisory Mechanism) (Pacific Small Island Developing States, 2025). The proposal mandates the Advisory Mechanism to “…inter alia, consider and make recommendations to the Conference and its subsidiary bodies on matters pertaining to the rights-based use of the traditional knowledge of Indigenous Peoples and of local communities in the implementation of the Agreement.” (Pacific Small Island Developing States, 2025). While the proposal is still being considered, should the Advisory Mechanism be established, it could lead the efforts with the secretariat and other stakeholders in the design and implementation of safeguards on the use or prohibition of AI on or information associated with traditional knowledge.

8 Conclusion

The Mechanism is considered the key institutional arrangement to support transparency and implementation across the Agreement's four substantive elements, including CB&TMT. In the Consolidated Draft Study on the Mechanism, using AI as an overlay to support improved user experience of the Mechanism is mentioned. AI-assisted tools could reduce administrative burden for CB&TMT associated needs assessments, summaries, and information retrieval, but only if they are deployed as decision-making support and remain subject to intergovernmental accountability and human control (OECD, 2019a,b; Jobin et al., 2019; ISO/IEC 42001, 2022). Therefore, the relevant policy question on the use of AI for the Mechanism is not whether AI is or not used but under what minimum conditions its use is consistent with implementation of the Agreement and trusted by the Parties and other key stakeholders, such as Indigenous Peoples and local communities. Further, the Mechanism has an opportunity to go beyond minimum conditions and lead the way in setting new standards for best practices. What such standards may be, is a question that will need to be addressed as the Mechanism, should it adopt the use of AI, reviews and assesses next steps.

On actual and potential risks for incorporating AI to the Mechanism, ensuring that the following three minimum requirements discussed above are met is essential, namely, 1. human approval for consequential outputs, 2. transparency and traceability, and 3. data access and governance controls. Such requirements are current best practices for the rapidly evolving IT industry, as adapted to the unique features of the Mechanism. Ocean Matcher, as an illustrative example that uses AI for matchmaking for funding associated with the ocean, may provide a good reference point for some of the practices and lessons learned on designing and implementing safeguards on the use of AI.

If designed and implemented well, a streamlined approach using AI could provide a better platform for fostering regional and multi-regional coordination and collaboration. AI could support such partnerships by conducting otherwise tedious administrative tasks (e.g., sifting through documents, organizing data and summarizing documents and information, etc.) for better cooperation and improving communication among partners. That said, procedural questions remain as to how such regional cooperation may work within the Agreement's CB&TMT requirements.

Further, on perceived risks, engagement with administrators and users will be essential to ensure that they are adequately addressed. In that regard, awareness raising, including workshops and information papers may be helpful. Finally, co-development of protection of traditional knowledge led by Indigenous Peoples and local communities will be particularly important, if traditional knowledge will be handled by the Mechanism. Should the Advisory Mechanism be established as proposed by the Pacific Small Island Developing States, it could lead in providing advice and coordinating with the secretariat and other stakeholders so that the Conference of the Parties can make an informed decision on a fit-for-purpose safeguards for the AI tool or to prohibit the use of AI on and information associated with traditional knowledge.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Sebastian Christoph Alexander Ferse, IPB University, Indonesia

Reviewed by: Yen-Chiang Chang, Dalian Maritime University, China

Lipianina-Honcharenko Khrystyna, West Ukrainian National University, Ukraine

Footnotes

Agreement Under the United Nations Convention on the Law of the Sea on the Conservations and Sustainable use of Marine Biological Diversity of Areas beyond National Jurisdiction, Adopted 19 June 2023, C.N.203.2023. Treaties-XXI.10. (BBNJ Agreement or Agreement).

The United Nations Treaty Collection, https://treaties.un.org/Pages/ViewDetails.aspx?src=TREATY&mtdsg_no=XXI-10&chapter=21&clang=_en, visited on 18 March 2026.

BBNJ Agreement, Part IV.

BBNJ Agreement, Part III.

BBNJ Agreement, Part II.

BBNJ Agreement, Part V.

BBNJ Agreement, Art. 51.7. Note that there will likely be different levels of access, (e.g., secretariat, Party's administrators, Indigenous Peoples and local communities, general users for open information and data, etc.), see Consolidated Draft Study on the Operationalization of the Mechanism.

BBNJ Agreement, Art. 51.

Pursuant to a notice from the Office of Legal Affairs of the United Nations and the Division for Ocean Affairs and Law of the Sea, until the Mechanism is operationalized, information exchange will be handled by the interim secretariat. https://www.un.org/bbnjagreement/sites/default/files/2026-02/BBNJNotification2026001.pdf, last visited, 3 May 2026.

https://oceandecade.org/ocean-decade-foundations-dialogue/, last visited 24 April 2026.

https://www.oceanmatcher.com/, last visited 24 April 2026.

BBNJ Agreement, Art. 42.4.

BBNJ Agreement, Art. 33 and Annex II.

BBNJ Agreement, Art. Annex II.

BBNJ Agreement, Art. 44.3.

BBNJ Agreement, Art. 41.2., 42.1, 42.4, 43.3, 43.5.

BBNJ Agreement, Art. 41.2.

BBNJ Agreement, Art. 45.3.