AI on BBNJ High Seas Biodiversity Agreement for Compliance teams at Renewables & Clean Energy firms in international jurisdictions

Executive Summary

The BBNJ High Seas Biodiversity Agreement establishes environmental impact assessment requirements for activities conducted in areas beyond national jurisdiction — a framework with direct relevance to Renewables & Clean Energy firms developing offshore projects, subsea cable routes, or deep-sea infrastructure in international waters. Compliance teams at these firms need accurate guidance on when EIA obligations are triggered, including the precise legal threshold and the correct article reference within the Agreement.

Testing AI tools on this regulation revealed a material hallucination: the AI replaced the Agreement's precautionary "may have more than a minor or transitory effect" screening standard with the narrower "likely to have" formulation, and simultaneously cited the wrong article as the source of that obligation. That combination of threshold misstating and article misidentification represents a compounded error that could lead a Compliance team to conclude that a planned activity does not require assessment when the Agreement's actual text would require one.

How AI gets this regulation wrong

The AI errors found on the BBNJ Agreement share a pattern of confident but legally consequential rule-restatement: AI tools quietly shifted the wording of treaty obligations in ways that look plausible on first reading but alter the practical legal standard. In this case, the failure involved both a misstated screening threshold — one that raises the bar for when an environmental impact assessment is required — and an incorrect article citation, so that independent verification against the treaty text becomes harder rather than easier. The table below breaks down how that failure mode manifests.

What that means for your team

The risk arising from AI errors on the BBNJ Agreement falls squarely in the regulatory enforcement category — a direct concern for Compliance functions at Renewables & Clean Energy firms that consult AI tools to scope EIA obligations for offshore or international-waters projects. For firms operating under the Agreement's emerging framework, an inaccurate threshold translates into the firm's exposure to having conducted a regulated activity without the required prior assessment. The table below maps that enforcement risk through the Compliance lens.

When this affects your department

Compliance teams at Renewables & Clean Energy firms in international jurisdictions encounter the BBNJ Agreement when assessing whether planned offshore activities — subsea cable installations, floating offshore wind development, ocean current or tidal energy projects, or infrastructure supporting deep-sea resource extraction relevant to battery supply chains — require environmental impact assessments under international law. These teams may use AI tools to quickly establish the legal framework: what triggers an EIA, at what article in the Agreement the obligation sits, and how the precautionary screening threshold is worded.

Policy briefs, regulatory mapping matrices, and due-diligence memos drafted with AI assistance carry these answers forward into board-level decisions, legal sign-off, and project development planning.

If the AI's answer contains a misstated trigger — replacing the Agreement's precautionary "may have more than a minor or transitory effect" standard with a narrower probability-based formulation, and attributing it to the wrong article — the Compliance team may advise project developers that a planned activity does not reach the assessment threshold when the Agreement's actual text would require one.

That error embedded in internal compliance documentation, an environmental governance policy, or a regulatory submission is difficult to correct after the fact, particularly as the Agreement moves toward broader ratification and its EIA requirements become more actively monitored by multilateral bodies.

The reputational dimension is equally significant for firms in this sector. Renewables & Clean Energy companies frequently compete for project rights and financing on the strength of their environmental due-diligence credentials. An AI-derived misstatement of international EIA obligations — if it surfaces in a project prospectus, sustainability disclosure, or engagement with a port-state or international body — can undermine investor and counterparty confidence at exactly the moment when the firm needs to demonstrate that its international operations meet the highest environmental standards, not merely the minimum.

The findings at a glance

The table below summarises the finding from our testing of AI tools on the BBNJ High Seas Biodiversity Agreement, covering the question area examined, the type of error the AI produced, and the risk category it creates for Compliance teams at Renewables & Clean Energy firms operating internationally.

Aggregate impact

The confirmed hallucination on the BBNJ Agreement identifies a precise and consequential point of failure: the EIA screening threshold. AI tools restated the treaty obligation using different language — substituting "likely to have" for the Agreement's actual "may have more than a minor or transitory effect" formulation. That substitution is subtle in form but significant in legal effect. The Agreement's actual standard is precautionary: a planned activity must undergo assessment if there is any real possibility that its impact exceeds the threshold, including where effects are unknown or poorly understood.

The AI's version converts that into a probability-based test, implying that assessment is only required when harm is considered probable. For Compliance teams advising on project design or regulatory strategy, those are materially different positions.

The article misidentification compounds the problem in a specific way. When a Compliance team attempts to verify the AI's answer by checking the treaty text, they are directed to Article 30 — not Article 27, where the screening provision is actually located. If the team searches Article 30 and does not find matching language, the natural inference may be that the AI summarised adjacent provisions rather than that it cited the wrong article altogether.

That dynamic makes the error self-concealing: the team's verification effort is directed at the wrong place, and the misstatement may survive internal review precisely because a check was performed.

For Renewables & Clean Energy firms, the systemic risk is that EIA threshold errors propagate into exactly the documents where they are hardest to catch later — environmental due-diligence frameworks, project feasibility assessments, and regulatory correspondence — at a stage when the Agreement is still developing its enforcement architecture and firms are establishing their compliance baselines. An error embedded at that stage can shape internal policy for several project cycles before it is detected, if it is detected at all.

What your team should do

The default position for Compliance teams using AI tools on the BBNJ Agreement should be to treat AI-generated treaty summaries as a research starting point rather than a definitive statement of the text. The Agreement's EIA framework — particularly the screening threshold and the article in which it sits — should always be verified directly against the treaty text, which is publicly available through the United Nations Treaty Collection portal.

Given the confirmed error pattern, teams should apply heightened scrutiny whenever AI output describes the trigger for an EIA obligation or cites specific article numbers, since both elements were misrepresented in testing.

In practice, this means building an explicit verification step into any internal workflow where AI is used to scope BBNJ obligations. When AI-generated content is used to draft sections of environmental due-diligence checklists, regulatory mapping documents, or internal training materials on the Agreement, a designated reviewer should cross-reference the AI's stated threshold language and article citations against the treaty text before the document is circulated or relied upon. This step is especially important where the output will inform a determination about whether a planned offshore activity requires an EIA — a determination with direct legal consequence for the firm.

AI tools remain useful for Compliance workflows on the BBNJ Agreement in contexts where precision on specific threshold wording is not required: gathering background on the Agreement's scope and negotiating history, identifying broadly which chapters address EIA obligations, or flagging which provisions are likely to apply to a new type of offshore energy project.

For any use case that turns on the specific legal standard — above all the precautionary "may have more than a minor or transitory effect" formulation and its location in the Agreement — teams should use AI output as a prompt for further research rather than as a final answer.

How RLB Can Help

RegLeg's published Hallucination Research gives Compliance teams at Renewables & Clean Energy firms a concrete pre-flight check before acting on AI-generated regulatory output. The research documents, by regulator and regulation, the specific question types where AI tools have demonstrably produced wrong answers — misquoted thresholds, fabricated cross-references, outdated permit conditions — so your team can identify whether a workflow is touching territory where failure rates are elevated before a decision is made, not after.

Beyond the published findings, RLB works with Compliance functions directly to map which AI-supported workflows in your firm carry the highest hallucination exposure. Renewables and clean energy businesses face a particularly complex regulatory surface: environmental permitting, grid-access obligations, project-finance disclosure requirements, and evolving carbon-market rules often sit across multiple jurisdictions simultaneously, each with its own amendment cadence. A bespoke regulator deep-dive scopes that landscape against the failure-mode patterns RLB has documented, so your team has a prioritised view of where AI assistance is relatively safe and where human verification steps remain essential.

RLB also offers a confidential review of your firm's existing AI-use policy against RegLeg's failure-mode catalogue, with prioritised remediation recommendations framed around your actual workflows rather than generic AI-governance checklists. For teams looking to build durable internal capability, RLB can develop training material and CPD-aligned content your Compliance staff can use directly — practical, evidence-based guidance on how to apply AI tools to regulatory research while recognising and managing the risk of hallucinated output.