Executive Summary
The BBNJ High Seas Biodiversity Agreement — the first international treaty governing the conservation and sustainable use of marine biological diversity in areas beyond national jurisdiction — introduces mandatory environmental impact assessment (EIA) obligations that directly intersect with upstream and deep-sea activities undertaken by Oil & Gas firms operating internationally. ESG & Sustainability teams at those firms are responsible for mapping these obligations into internal governance frameworks, supply-chain due-diligence processes, and regulatory reporting.
Across the question set we put to AI tools on this regulation, AI assistants produced an incorrect answer on the core EIA screening threshold — the precise legal test that determines whether a planned activity must undergo formal assessment before proceeding. The failure took a specific and consequential form: the AI raised the bar for triggering an EIA by substituting a higher-certainty standard for the agreement's deliberately precautionary one, while simultaneously citing the wrong article as authority.
A team that relied on this answer without independent verification could design internal screening processes that systematically exempt activities the agreement requires to be assessed.
How AI gets this regulation wrong
The pattern of AI error on the BBNJ Agreement is subtle but structurally significant: AI tools stated rules with apparent confidence, only to acknowledge uncertainty when pressed — a dynamic that makes the error harder to catch during ordinary research workflows. In the finding documented here, the AI quietly rewrote a foundational legal threshold, replacing the treaty's precautionary trigger with a more demanding standard of proof, and anchored that misstated rule to the wrong article. The table below breaks down this failure mode and how it manifested.
| AI's Failure Mode | Count | Affected findings |
|---|---|---|
| Exposed Fabrication | 1 | Finding#1 |
What that means for your team
For ESG & Sustainability teams at Oil & Gas firms, the risks arising from AI errors on the BBNJ Agreement concentrate squarely in regulatory enforcement exposure — the possibility that internal screening decisions made on the basis of a misstated legal standard leave the firm on the wrong side of treaty obligations. Because the agreement sits within international public law and is enforced through state-party obligations that cascade into domestic implementing legislation and permit conditions, a miscalibrated internal EIA threshold is not easily corrected after the fact.
The table below maps the specific risk categories that follow from the AI failures identified in this regulation.
| Risk Impact | Count | Affected findings |
|---|---|---|
| Regulatory enforcement | 1 | Finding#1 |
When this affects your department
ESG & Sustainability teams at Oil & Gas firms encounter the BBNJ Agreement in a growing range of operational and governance contexts. As ratification progresses and state parties begin transposing treaty obligations into national licensing regimes, firms with exploration, extraction, or subsea infrastructure activities in international waters — including deep-sea pipeline routes, seabed surveys, and offshore platform positioning — need to understand which planned activities trigger the treaty's EIA requirements before regulatory applications are filed or project timelines are set.
AI tools are frequently consulted at the early scoping stage, when the team is deciding whether to commission a formal EIA at all, and the threshold question — does this activity clear the screening bar? — is exactly where an incorrect AI answer does the most damage.
Beyond project-level scoping, the BBNJ Agreement features heavily in investor-facing ESG disclosures, sustainability frameworks, and international policy monitoring. Firms preparing TCFD-aligned or TNFD-aligned disclosures, or those responding to investor questionnaires about biodiversity risk, often task ESG teams with summarising relevant treaty obligations. A summary built on AI-generated content that misstates the EIA trigger — or cites the wrong article of the agreement as authority — embeds a structural error into documents that may be relied upon by investors, insurers, and counterparties.
The stakes compound because the agreement's EIA provisions are explicitly precautionary: the drafting is designed to capture activities whose effects are unknown or poorly understood, not just those already predicted to cause harm. An AI answer that narrows this threshold creates a compliance blind spot precisely where the treaty most intends to operate.
If an activity proceeds without the required assessment because the firm's internal screening tool was calibrated to a higher standard, the firm may face treaty non-compliance findings, domestic enforcement action under implementing legislation, or reputational consequences when the gap is identified during due diligence by lenders or acquirers.
The findings at a glance
The table below summarises the finding arising from AI testing on the BBNJ Agreement that is relevant to ESG & Sustainability teams at Oil & Gas firms, including the nature of the AI error and its primary risk category.
| # | Finding title | Type | Citation ID |
|---|---|---|---|
| 1 | EIA screening threshold misstated with wrong article citation | Hallucination | RLB-F-INT-UNTC-BBNJ-HIGH-SEAS-BIODIVERSITY-AGREEMENT-2023-Q001 |
Aggregate impact
The finding on the BBNJ Agreement reveals an error pattern that is particularly hazardous for ESG & Sustainability teams: the AI's mistake was not a wholesale fabrication of a non-existent rule, but a precise alteration of an existing one. By replacing "may have more than a minor or transitory effect" — the treaty's actual screening threshold — with "likely to have more than a minor or transitory effect," the AI shifted the standard from a precautionary, uncertainty-inclusive test to a probability-of-harm test. That single-word substitution moves the legal boundary in a direction that systematically favours non-assessment.
Compounding this, the AI cited the wrong article as the source of the threshold, meaning a team that tried to verify the answer by checking the cited article would find plausible but misattributed text.
The error clusters around the EIA screening provisions, which are the operational core of the agreement for any firm with activities in international waters. This is not a peripheral or technical clause — it is the threshold that determines whether a project must pause for environmental review before proceeding. For Oil & Gas firms, that threshold is directly relevant to subsea surveys, pipeline route assessments, and any exploratory activity in the high seas.
An internal EIA screening tool or decision matrix built on the AI's misstated standard would fail to flag activities that the agreement requires to be assessed, and that failure would be invisible to the team unless they had independently verified the treaty text.
The systemic risk for firms operating internationally is that AI errors of this type — confident, source-cited, but structurally wrong — tend to survive internal review processes that treat AI-generated summaries as a starting point rather than a claim requiring verification. Where the ESG & Sustainability function is also the internal expert on this regulation, there may be no independent check before the misstated threshold is embedded into governance documents or project approvals.
What your team should do
The default position for ESG & Sustainability teams consulting AI tools on the BBNJ Agreement should be: treat AI-generated summaries of specific legal thresholds as a starting point for locating the relevant provision, not as a reliable statement of what that provision says. The finding here demonstrates that AI tools can misstate a threshold while citing a real source, and can assign that threshold to the wrong article — both errors that survive a surface-level plausibility check.
For any question about whether a planned activity triggers an EIA obligation under the agreement, the definitive answer must be drawn from the treaty text itself, available through the United Nations Treaty Collection at treaties.un.org, and preferably reviewed by legal counsel familiar with international environmental law.
Practical safeguards for the team's workflow include: (1) building internal EIA screening tools and decision matrices from the treaty text rather than from AI-generated summaries of it; (2) citing specific articles directly in any governance document that states a legal threshold, so that future reviewers can verify the source; and (3) when preparing investor disclosures or policy summaries that reference the agreement's EIA provisions, having the specific article cross-checked by someone with direct access to the treaty text before the document is finalised.
Where the firm's legal team has not yet developed internal guidance on the BBNJ Agreement, flagging this gap proactively is more protective than relying on AI to fill it.
AI tools remain useful for ESG & Sustainability teams working on the BBNJ Agreement in contexts that do not require precision on specific legal thresholds: generating a high-level overview of the agreement's structure and objectives for internal briefings, identifying which areas of the agreement are most relevant to the firm's operational footprint, or drafting background sections of stakeholder communications that will be reviewed and edited before publication. The risk is concentrated in exactly the kind of precise threshold question that is most tempting to delegate to AI — and most consequential if the AI's answer is wrong.
How RLB Can Help
RegLeg's published Hallucination Research gives ESG & Sustainability teams at oil and gas firms a ready-made pre-flight check before acting on AI-assisted regulatory analysis. The research documents, by regulator and rule, the specific ways AI tools have misrepresented disclosure thresholds, misquoted transition-plan requirements, and conflated jurisdiction-specific standards — exactly the failure modes that carry legal and reputational weight for a firm operating across multiple reporting regimes. Using this as a standing reference, your team can identify which AI outputs warrant closer scrutiny before they inform a compliance position or a board-level disclosure decision.
Beyond the published research, RLB works with ESG & Sustainability functions to map the hallucination exposure specific to their operating model. Oil and gas firms face an unusually layered regulatory environment — voluntary frameworks sitting alongside mandatory disclosure rules, cross-border reporting obligations, and fast-moving guidance from bodies that AI tools have not yet reliably internalised. A bespoke deep-dive identifies which AI-supported workflows in your function carry the highest risk, whether that is screening emissions data against regulatory thresholds, drafting narrative disclosure, or monitoring for regulatory change across jurisdictions.
RLB also provides 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 immediately rather than a generic risk register.
For teams looking to build durable capability, RLB produces training material and CPD-aligned content that can be delivered internally. The focus is practical: how to read AI output critically, which question types reliably surface unreliable responses in an ESG & Sustainability context, and how to maintain appropriate oversight as AI tools become more embedded in day-to-day regulatory work. Content can be calibrated to the seniority mix of your team, from analysts running initial regulatory screens through to senior professionals accountable for signed disclosures.