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

Executive Summary

The BBNJ High Seas Biodiversity Agreement, adopted under the United Nations Convention on the Law of the Sea, establishes a binding international framework for the conservation and sustainable use of marine biodiversity in areas beyond national jurisdiction — including environmental impact assessment obligations that apply directly to high-seas activities. For Compliance teams at Maritime & Shipping firms operating across international jurisdictions, understanding precisely when those EIA obligations are triggered is a core regulatory task.

Across the question we tested on this regulation, AI tools produced an incorrect answer: the EIA screening threshold was misrepresented in a way that would lead a Compliance team to believe a narrower set of activities requires assessment than the Agreement actually mandates. The failure took the form of confident misinformation — and when the AI was challenged, it acknowledged uncertainty it had not disclosed in its original response.

How AI gets this regulation wrong

The table below details how AI tools went wrong when answering questions about this regulation. The predominant pattern here is confident misinformation: the AI provided a definitive-sounding answer that subtly but materially misrepresented the rule — and only backed away from that answer when pressed. This type of failure is particularly hazardous because it is indistinguishable, at face value, from a correct response.

What that means for your team

The table below maps each finding to its practical risk category for your team. For this regulation, the risks cluster around regulatory enforcement exposure: if a Compliance function acts on a miscalibrated understanding of the EIA screening threshold, the firm may conduct or authorise high-seas activities without the assessments the Agreement requires — placing it in breach of an international treaty obligation. That enforcement exposure is the primary lens through which to read these findings.

When this affects your department

Maritime & Shipping firms with operations in international waters — whether running deep-sea cable-laying, subsea resource surveys, transoceanic freight routes through ecologically sensitive corridors, or research vessel charters — are increasingly within scope of the BBNJ Agreement's environmental impact assessment regime. Compliance teams are typically the first internal function asked to map whether a planned activity crosses the treaty's EIA trigger. That mapping exercise often begins with a research question put to an AI assistant: what is the legal test, which article governs it, and at what point does an assessment become mandatory rather than advisory?

If the AI's answer to that question is wrong — and specifically if it substitutes a higher-bar "likely to have" standard for the Agreement's actual "may have more than a minor or transitory effect, or effects unknown or poorly understood" threshold — the Compliance team's internal advice to the business will systematically undercount the activities that require assessment. A route survey, a seabed mapping exercise, or a novel high-seas operation that genuinely sits within scope of the EIA obligation would be waved through without the required process.

The consequences for the firm are not merely technical. The BBNJ Agreement is an international treaty with State-party enforcement mechanisms, and flag-state, port-state, and coastal-state authorities may all have a role in holding shipping firms to account for EIA non-compliance. A firm that relied on an AI-generated threshold analysis — and whose internal policies, board submissions, and regulatory filings reflect that miscalibrated standard — faces not only direct enforcement risk but the harder problem of having to unwind compliance infrastructure built on a wrong foundation.

The findings at a glance

The table below summarises each finding in this cell — what was asked, what the AI got wrong, and the risk category it creates for a Compliance team at a Maritime & Shipping firm.

Aggregate impact

The single finding in this cell points to a precise and consequential error: the AI replaced the BBNJ Agreement's precautionary EIA trigger with a materially higher-bar standard. The Agreement's text — as confirmed by the United Nations Treaty Collection — requires an environmental impact assessment where an activity "may have more than a minor or transitory effect on the marine environment, or the effects of which are unknown or poorly understood." The AI instead framed the threshold as activities "likely to have" such an effect, and cited the wrong article as the source of the screening provision.

Both errors run in the same direction: they make the EIA obligation appear narrower and more certain than it actually is.

For Compliance teams at Maritime & Shipping firms, this pattern of error is systematically dangerous. The BBNJ Agreement's precautionary framing is deliberate: the "may have" and "unknown or poorly understood" language is designed to capture activities where the environmental consequences are not yet clear, not just those where harm is probable. A Compliance function that uses a "likely to have" standard will produce assessments that exclude exactly the class of uncertain or novel activities the Agreement was designed to catch — which is precisely the category that includes many modern high-seas maritime operations.

The article misattribution compounds the analytical problem. When Compliance teams cite specific treaty provisions in internal policies, board papers, or submissions to flag-state or port-state authorities, a wrong article reference is not merely a formatting error — it signals that the team is working from a secondary or reconstructed account of the text rather than the treaty itself. That erodes the credibility of the compliance position and, in an enforcement context, may suggest the firm did not exercise the care required.

What your team should do

The default position for Compliance teams using AI tools on the BBNJ Agreement should be: AI is not a reliable source for the precise legal standard that triggers an EIA obligation, nor for the specific article reference that governs it. Both elements — the threshold wording and the article number — are exactly the kind of granular treaty detail where AI tools have been shown to confidently produce wrong answers. Any AI-generated summary of the EIA trigger should be treated as a starting point for verification, not a final answer.

Practical safeguards for this regulation are straightforward. The authoritative text is the BBNJ Agreement itself, available directly through the United Nations Treaty Collection at treaties.un.org. Article verification takes minutes against the primary source and should be a non-negotiable step before the threshold standard is embedded in internal policies, compliance frameworks, or communications with flag-state or port-state authorities. When in doubt about whether a planned activity falls within scope, the Agreement's "unknown or poorly understood" limb of the EIA trigger is a signal that the precautionary framing should be applied — not that the activity is outside scope.

Where AI tools are genuinely useful for Compliance teams working on the BBNJ Agreement is in background orientation: understanding the overall structure of the treaty, the relationship between the EIA provisions and the broader area-based management framework, or summarising public commentary and guidance from United Nations bodies. These tasks draw on information that is widely reproduced and less dependent on article-level precision.

The risk emerges when AI output moves from orientation to authoritative legal characterisation — and on this regulation, that line is crossed as soon as the EIA threshold or its article reference appears in AI-generated text without independent verification.

How RLB Can Help

RegLeg's published Hallucination Research gives Compliance teams at Maritime & Shipping firms a ready-made pre-flight check before acting on AI-generated output. Each research entry documents specific instances where AI tools have mischaracterised regulatory obligations — including on port state control requirements, flag state certification rules, and cross-border trade compliance — so your team can identify the regulatory areas where a second check is warranted before relying on an AI-assisted answer.

Where the published research surfaces a pattern relevant to your firm, RLB can carry that into a bespoke regulator deep-dive: a structured mapping of which AI-supported workflows in your Compliance function carry the highest hallucination exposure. For Maritime & Shipping Compliance this typically spans vessel classification queries, cargo documentation obligations, sanctions screening workflows, and multi-jurisdictional port entry requirements — areas where AI tools draw on dense, frequently amended regulatory text and where a confidently wrong answer carries direct operational and legal consequence.

RLB also offers a confidential review of your firm's existing AI-use policy against our failure-mode catalogue, returning a prioritised remediation list your team can act on without disclosing internal processes. Alongside that review, we can provide training material and CPD-aligned content your Compliance team can use internally — grounding staff in the specific failure patterns most relevant to maritime regulatory practice and giving them a repeatable framework for calibrating AI output as part of day-to-day compliance work.