AI on BBNJ High Seas Biodiversity Agreement for Legal teams at Maritime & Shipping firms in international jurisdictions

Executive Summary

The BBNJ High Seas Biodiversity Agreement — formally the Agreement under UNCLOS on the Conservation and Sustainable Use of Marine Biological Diversity of Areas Beyond National Jurisdiction — establishes the first binding international framework for environmental governance of the high seas, introducing mandatory environmental impact assessments, area-based management tools, and a Conference of the Parties with authority over marine protected areas in international waters. For Legal teams at Maritime & Shipping firms operating in international jurisdictions, understanding its precise obligations and governance structure is critical: fleet operations, voyage planning, and port-state compliance programmes all intersect directly with this treaty's reach.

Across the two questions we put to AI tools on this regulation, AI got both wrong — mischaracterising the threshold that triggers an environmental impact assessment and misattributing the article that limits the Conference of the Parties' authority over vessel transit rights in shipping lanes. In both cases the AI produced a confident, plausible-sounding answer that a Legal team could easily mistake for accurate analysis, yet each answer contained a material legal error that would distort the advice or documents built from it.

How AI gets this regulation wrong

AI tools struggled with the BBNJ Agreement in two distinct ways: inventing a different article number for a well-defined provision, and quietly raising the legal threshold that triggers an obligation — making a wider duty appear narrower than the treaty actually requires. Both failures share a common thread: the AI produced responses that sounded authoritative and were structurally coherent, yet contained errors that only surface when the answer is checked against the treaty text itself.

What that means for your team

For a Legal team at a Maritime & Shipping firm, errors in this regulation carry a common downstream risk: a wrong deliverable — advice, policy text, or a legal opinion built on a misstatement of what the treaty actually says. Both findings fall into this category, and the consequences range from incorrectly scoped environmental due-diligence to a flawed assessment of the firm's exposure to area-based management restrictions in international shipping routes.

When this affects your department

Legal teams at Maritime & Shipping firms are among the most directly affected practitioners under the BBNJ Agreement. The treaty's environmental impact assessment regime applies to a broad range of planned high-seas activities, and legal counsel must be able to advise on whether a proposed voyage, offshore operation, or transoceanic infrastructure project crosses the EIA screening threshold — and if so, what procedural obligations follow.

Teams will consult AI tools when drafting internal compliance frameworks for flag-state ratification planning, producing guidance notes for operational and crewing functions, or advising the commercial desk on whether a new trade route or activity type is likely to attract regulatory scrutiny under the Agreement.

The treaty's area-based management tool provisions are equally live for shipping lawyers. The BBNJ Conference of the Parties has authority to designate marine protected areas in international waters, and the interaction between that authority and established transit rights under UNCLOS — including rights of innocent passage, transit passage, and freedom of navigation — is an area of active legal uncertainty. Legal teams advising on fleet strategy, voyage chartering, or regulatory risk in specific ocean corridors need to understand exactly where the COP's power to restrict vessel movement ends, and which article of the Agreement governs that boundary.

If an AI-generated analysis is wrong on either of these points, the consequences for the firm compound quickly. An incorrectly scoped EIA screening threshold can cause a project team to bypass an assessment that the treaty actually requires, exposing the firm to enforcement action by flag states, flag-state deregistration risk, or liability in contractual indemnity chains where environmental compliance is warranted.

A misidentified article governing COP authority can produce a legal opinion that overstates or understates the firm's exposure to transit restrictions in a designated zone — either causing the firm to over-invest in contingency routing or, more dangerously, to operate under the false assumption that a restriction cannot lawfully reach it.

The findings at a glance

The table below summarises each finding on the BBNJ Agreement for this audience — the question asked, what the treaty actually says, and how the AI's answer diverged from it.

Aggregate impact

Both findings on the BBNJ Agreement cluster on the same underlying problem: the AI's answers were structurally plausible but legally incorrect in ways that only become visible when the answer is checked against the treaty text at the article level. In one case, the AI substituted a higher-bar trigger for the EIA screening obligation — replacing the precautionary "may have" standard set out in the Agreement with the more demanding "likely to have" formulation, which is not what the treaty says.

In the other, the AI correctly identified the principle limiting the Conference of the Parties' authority over area-based management decisions but attributed it to the wrong article, conflating a general UNCLOS-relationship clause with the specific provision that governs COP decisions on marine protected areas. Both errors share a profile: the AI knew enough about the treaty to sound credible, but not enough to be accurate.

For a Legal team at a Maritime & Shipping firm, this pattern is particularly dangerous because the errors are not obviously wrong. A lawyer who receives the AI's EIA screening answer and does not verify the exact threshold language against the treaty text will carry the misstatement forward — into a compliance note, an internal training deck, or external advice to a commercial counterparty. Similarly, an opinion citing the wrong article for the non-undermining rule will still read coherently to a reader unfamiliar with the treaty's structure; the error only surfaces in adversarial challenge or enforcement.

The systemic risk is that both errors narrow the firm's perceived obligation or overstate the firmness of its legal position: the EIA threshold error makes a wider screening obligation appear less demanding than it is, and the article misattribution could produce an opinion that incorrectly treats the COP's authority as more constrained — or less constrained — than the specific provision actually provides. Firms that use AI-assisted legal research on the BBNJ Agreement without article-level verification against treaty text are exposed to both forms of error simultaneously.

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 structured first draft that identifies the relevant topic area, then verify every threshold, obligation trigger, and article reference against the treaty text before the analysis leaves the team.

The BBNJ Agreement is a recently adopted treaty with a relatively small body of secondary commentary, which means AI tools are more likely to draw on analogical reasoning from related instruments — such as UNCLOS or the London Protocol — or on pre-ratification academic literature that may not reflect the final treaty text. Neither is a reliable substitute for the agreement itself.

For the EIA screening question specifically, Legal teams should go directly to the relevant treaty articles and confirm the precise trigger language before advising on whether a planned activity requires an assessment. The difference between "may have" and "likely to have" as threshold standards is material in environmental law: it determines which activities fall inside the obligation and which can be lawfully excluded from formal assessment. AI tools are prone to smoothing over this distinction by defaulting to more familiar formulations from national environmental legislation or other international instruments.

The same discipline applies to area-based management questions: any opinion on the COP's authority — and its limits — should cite the article that specifically governs COP decisions on marine protected areas, not the general provision addressing the Agreement's relationship with existing legal frameworks.

AI tools remain useful for Legal teams navigating the BBNJ Agreement in lower-stakes contexts: generating a first-pass summary of treaty structure, identifying which chapters cover EIA versus ABMT versus benefit-sharing, or drafting a list of questions for a more detailed review. Where AI is most reliable is on the treaty's architecture — the broad sequence of obligations and governance institutions — rather than on the precise legal standards that determine whether a specific obligation is engaged. Use AI to map the terrain; use the treaty text to draw the lines.

How RLB Can Help

RegLeg's published Hallucination Research gives the Legal team at a Maritime & Shipping firm a ready pre-flight check before relying on AI-assisted output for regulatory questions. The research catalogues, by regulator and rule-set, the specific failure modes AI tools exhibit when asked about flag-state requirements, port-state control obligations, cargo liability frameworks, and cross-border sanctions regimes. Legal teams can consult the findings before drafting advice, reviewing contracts, or preparing regulatory submissions — treating the research as a structured second opinion on where AI tools have demonstrably produced incorrect answers on the very regulations in scope.

For firms that have moved beyond ad hoc AI use into structured workflows, RegLeg offers bespoke regulator deep-dives scoped to the Legal function's actual workload. These engagements map which AI-supported tasks — sanctions screening queries, ISM Code compliance checks, charter-party clause interpretation, incident-reporting deadline calculations — carry the highest hallucination exposure across the relevant international regulatory stack, and produce a prioritised risk register the team can act on immediately. Where a firm's Legal team operates across multiple flag and port-state jurisdictions, the deep-dive can be structured to reflect that multi-regulator exposure directly.

RegLeg can also conduct a confidential review of the firm's existing AI-use policy against its failure-mode catalogue, identifying gaps between current guardrails and the hallucination patterns the research has documented in practice, with a prioritised remediation roadmap. For teams with internal training obligations or CPD requirements, RegLeg develops training material and CPD-aligned content drawn directly from the published research — giving Legal professionals concrete, evidenced examples of AI failure in maritime and shipping regulatory contexts that are credible with both compliance committees and external counsel.