AI on BBNJ High Seas Biodiversity Agreement for Legal teams at Pharmaceuticals firms in international jurisdictions

Executive Summary

The BBNJ Agreement — the United Nations treaty on the conservation and sustainable use of marine biological diversity in areas beyond national jurisdiction — introduces a new international framework governing access to marine genetic resources (MGRs) from the high seas and the fair sharing of benefits derived from their utilisation, including from digital sequence information (DSI). For Legal teams at Pharmaceuticals firms operating across international jurisdictions, the agreement is directly material: many firms hold or are building programmes that rely on high-seas biological samples and the genomic data extracted from them.

Across the questions we tested, AI assistants produced incorrect answers on both of the substantive questions put to them in this cell. The errors were not minor misstatements — in one case AI tools inverted the agreement's retroactivity rule, asserting that benefit-sharing obligations apply by default to pre-agreement collections, when the treaty text says the opposite; in the other, AI tools cited the wrong article number for the DSI benefit-sharing provision.

Both failures expose a Legal team to the risk of producing a wrong deliverable — internal policy guidance, due-diligence advice, or external compliance documentation — built on a materially false reading of the treaty.

How AI gets this regulation wrong

Across both findings in this cell, the dominant pattern is confident fabrication: AI tools stated rules with apparent authority, then — when challenged — acknowledged they were not certain of the underlying text. In both instances the AI produced answers that were not simply incomplete but structurally wrong, inverting a treaty default or anchoring a correct principle to the wrong provision. The table below sets out the breakdown of how AI tools got this regulation wrong.

What that means for your team

For a Legal team at a Pharmaceuticals firm, both failure modes map directly to the same risk category: the team produces a deliverable — internal guidance, a supplier due-diligence report, a regulatory mapping exercise — that states the law incorrectly, and that document then shapes commercial or operational decisions built on a false foundation. The risk is not merely reputational; where the deliverable informs a transaction or a regulatory filing, the downstream consequences can include regulatory non-compliance, contractual exposure, and remediation costs. The table below sets out the risk impact breakdown for this regulation.

When this affects your department

A Legal team at a Pharmaceuticals firm in international jurisdictions will encounter the BBNJ Agreement in several practical workflows. The most common is regulatory mapping for R&D programmes: any programme that sources biological material from marine environments in areas beyond national jurisdiction, or that processes or licences genomic data derived from such material, needs to understand what the agreement requires in terms of access procedures, benefit-sharing payments, and disclosure obligations.

Legal teams are also frequently asked to advise on the treatment of legacy collections — samples gathered before the treaty's entry into force — because the question of whether existing biobanks or data repositories fall within the agreement's scope directly affects whether benefit-sharing obligations attach.

A second common scenario is supplier and partner due diligence. Pharmaceutical firms that do not collect marine samples directly may nonetheless work with contract research organisations, biotechnology partners, or academic collaborators that do. Legal teams are asked to assess whether those upstream relationships expose the firm to benefit-sharing obligations or reporting requirements, and to draft or review the relevant contractual clauses. Questions about the scope of DSI coverage — which article governs it, what it requires, and how it interacts with access and benefit-sharing mechanisms — arise directly in this context.

If a Legal team reaches for AI tools to answer these questions and receives incorrect guidance, the consequences propagate quickly through the firm. A policy memo stating that pre-agreement collections are retroactively subject to benefit-sharing will cause the business to over-comply, potentially requiring renegotiation of supplier contracts and triggering unnecessary reporting burdens. Conversely, a memo that wrongly anchors the DSI benefit-sharing rule to the wrong treaty provision may cause the firm to misread the obligation's scope — for example, misidentifying which activity triggers the obligation — leading to under-compliance and potential exposure when national implementing legislation is enacted.

The findings at a glance

The table below summarises the two findings from this regulation that are relevant to Legal teams at Pharmaceuticals firms in international jurisdictions, including the question area, the nature of the AI's error, and the risk category each error falls into.

Aggregate impact

The two findings in this cell cluster around a single operational theme: the scope and trigger conditions of the BBNJ Agreement's benefit-sharing obligations. Both the retroactivity question and the DSI article-number question are the kind of foundational interpretive issues that a Legal team resolves early, before advice flows downstream to R&D, business development, or procurement. When AI tools get these foundational questions wrong, the error does not stay in one document — it becomes embedded in the firm's working assumptions about what the treaty requires.

The retroactivity error (Finding 1) is the more commercially significant of the two. AI tools tested on this question inverted the treaty's non-retroactivity default, asserting that benefit-sharing obligations apply to pre-agreement collections unless a Party expressly opts out. The actual text states the opposite: the regime applies only to resources collected and generated after the agreement's entry into force for each Party, with no opt-out mechanism needed because retroactive application is simply not the default.

For a Pharmaceuticals firm with legacy marine biobank holdings, or for one that has already licensed data from pre-entry-into-force collections, this inversion is not a technical quibble — it determines whether a substantial body of existing material is subject to new financial and procedural obligations.

The article-citation error (Finding 2) is more contained in isolation but carries compounding risk. An incorrect cross-reference in internal legal guidance means that when staff or external counsel verify the basis for the firm's position, they find a mismatch between the cited provision and its actual text. In a regulatory correspondence context — where treaty references are checked carefully by counterparties and regulators — this kind of error undermines the firm's credibility and may prompt scrutiny of related positions.

Taken together, the two findings show that AI tools are currently unreliable on the most basic interpretive questions about this treaty, and that reliance without verification creates a systemic risk of producing misdirected legal work product.

What your team should do

The default position for Legal teams using AI tools on the BBNJ Agreement should be: treat AI output as a first-draft research prompt, not as a source of treaty interpretation. The treaty is recent, its implementation is still developing across jurisdictions, and the secondary commentary available to AI tools is uneven in accuracy — as evidenced by the Pretextual sources cited in Finding 1, which discussed the agreement but misrepresented its operative provisions. Until national implementing legislation consolidates the framework, every AI-generated statement about the agreement's scope, obligations, and article structure requires verification against the treaty text itself.

For the specific risk areas identified in this cell, the practical safeguard is straightforward: before any internal document states a position on retroactivity or on which provisions govern DSI, a team member should read the relevant article directly from the treaty text as deposited with the United Nations Treaty Collection. Article 10(1) on temporal scope and Article 14(1) on DSI benefit-sharing are short, unambiguous provisions — primary verification takes minutes and eliminates both categories of error documented here.

Where the question is more complex — for example, how a specific Party's implementing legislation affects the firm's obligations — primary-text verification should be supplemented with specialist external counsel advice.

AI tools remain useful in the Legal workflow for orientation tasks that do not depend on precise article-level accuracy: generating an initial list of issues to investigate, drafting a framework for a regulatory mapping exercise, or summarising the high-level policy goals the treaty pursues. The risk is in using AI tools as a citation or interpretation authority.

Treating AI output as a starting point rather than a conclusion — and building a standing practice of primary-source verification into any BBNJ-related work product — removes the compounding risk that an unverified AI error travels from one document into the firm's broader position on this regulation.

How RLB Can Help

RegLeg's published Hallucination Research gives the Legal team at a Pharmaceuticals firm a practical pre-flight check before placing reliance on AI-assisted regulatory analysis. Because the research is organised by regulator and regulatory instrument, Legal counsel can look up the specific rules their team works with most — pharmacovigilance reporting obligations, marketing authorisation conditions, GMP requirements, cross-border distribution frameworks — and see precisely where AI tools have been shown to misstate statutory text, confuse version histories, or fabricate citation details.

Using that record as a standing reference costs nothing and takes minutes; it converts a vague concern about AI reliability into a concrete checklist against known failure modes.

Beyond the published research, RegLeg works with Pharmaceuticals Legal teams on bespoke regulator deep-dives that map which AI-supported workflows in the function carry the highest hallucination exposure. Regulatory affairs correspondence, internal compliance opinions, licence variation submissions, and pharmacovigilance signal assessments each carry different risk profiles, and the relevant regulators — EMA, FDA, MHRA, TGA and others — are not equally represented in AI training data. A structured deep-dive surfaces where the gap between an AI tool's apparent confidence and its actual accuracy is widest, allowing the team to concentrate human review where it matters most.

For firms that have already adopted AI use policies, RegLeg offers a confidential review of those policies against its failure-mode catalogue, returning a prioritised remediation plan that is practical to implement rather than aspirational in scope. Where the Legal team needs to build internal capability — whether for onboarding new staff, satisfying CPD requirements, or briefing business partners on appropriate use of AI tools — RegLeg can supply training material and CPD-aligned content tailored to the Pharmaceuticals regulatory environment and the specific audience within the Legal function.