Does the BBNJ Agreement apply to samples of marine genetic resources collected from the high seas before the agreement entered into force?
The model stated the opposite of the Agreement's default rule. The Agreement is prospective by design — it covers only post-entry-into-force collections — but the model described a retroactive regime with an opt-out, which inverts both the default and the opt-out structure. This appears to reflect the model reconstructing from commentary on earlier drafts of the text, where the retroactivity question was actively contested, rather than reading the final adopted provision.
This finding implicates the training data layer: the model appears to have learned the retroactivity rule from pre-adoption negotiating commentary that described an earlier draft rather than the final adopted text. The retrieval step did not correct this because the cited secondary source itself may contain the same error. Both training corpus curation and retrieval-source ranking need to weight post-adoption primary text over pre-adoption commentary for recently adopted instruments.
Does the BBNJ Agreement apply to marine genetic resources collected before the agreement entered into force, or does it operate prospectively only?
The model inverted the Agreement's default rule and also inverted the opt-out structure. The Agreement is prospective by default, with parties able to declare retroactive application if they choose. The model described the opposite: retroactive as default, with an opt-out. This is the same fundamental error observed in Claude Opus 4.7 with web search on the same question, which strongly suggests both models are drawing on commentary describing an earlier draft regime rather than the final adopted text.
This finding, alongside the Opus 4.7 retroactivity finding, strongly suggests a shared training-data origin: both models inverted the same rule in the same direction. The implication for your team is that the error is not a model-specific calibration problem — it is a corpus-level issue that will persist across model versions until the training data for this instrument is corrected. Targeted correction pairs anchored to the final adopted Article 10(1) text are likely the most efficient fix.
This is the highest-risk finding in the cell for Law Firms firms: the AI assistants we tested completely inverted the BBNJ Agreement's retroactivity rule, stating that the marine genetic resource and digital sequence information benefit-sharing regime applies retroactively by default to pre-entry-into-force collections — with an opt-out — when the correct position is the exact opposite (non-retroactive by default under Article 10(1)).
A Legal team acting on this error could advise a client in the marine biotech, pharmaceuticals, or food ingredient sectors to undertake unnecessary benefit-sharing compliance action for legacy collections, or could produce a transaction due-diligence report that mischaracterises the regulatory exposure of an acquisition target's existing MGR portfolio. Where the error travels into a client opinion that is relied upon in a completed transaction, the firm's professional indemnity position is exposed to the full extent of any resulting loss.
Each finding has a stable Citation ID (RLB-F-… for aggregated case-study findings, RLB-H-… for raw per-model hallucinations) — like a DOI, the ID always resolves to the canonical finding even if URLs change.
RegLeg Specialist Panel (2026). "Finding#2 — MGR retroactivity default inverted by AI — Law Firms × Legal — International / Multilateral." Citation ID: RLB-F-INT-UNTC-BBNJ-HIGH-SEAS-BIODIVERSITY-AGREEMENT-2023-Q003. RegLegBrief AI Hallucination Research, published 2026-06-03. https://reglegbrief.com/regulators/j1/int/untc/bbnj-high-seas-biodiversity-agreement-2023/sectors/law_firms/legal/finding/INT-UNTC-INT-001-BBNJ-HIGH-SEAS-BIODIVERSITY-AGREEMENT-2023-v1-003/
RegLeg Specialist Panel. (2026). Finding#2 — MGR retroactivity default inverted by AI [Hallucination finding RLB-F-INT-UNTC-BBNJ-HIGH-SEAS-BIODIVERSITY-AGREEMENT-2023-Q003]. RegLegBrief AI Hallucination Research. https://reglegbrief.com/regulators/j1/int/untc/bbnj-high-seas-biodiversity-agreement-2023/sectors/law_firms/legal/finding/INT-UNTC-INT-001-BBNJ-HIGH-SEAS-BIODIVERSITY-AGREEMENT-2023-v1-003/
RegLeg Specialist Panel, Finding#2 — MGR retroactivity default inverted by AI [RLB-F-INT-UNTC-BBNJ-HIGH-SEAS-BIODIVERSITY-AGREEMENT-2023-Q003], RegLegBrief AI Hallucination Research (June 03, 2026), https://reglegbrief.com/regulators/j1/int/untc/bbnj-high-seas-biodiversity-agreement-2023/sectors/law_firms/legal/finding/INT-UNTC-INT-001-BBNJ-HIGH-SEAS-BIODIVERSITY-AGREEMENT-2023-v1-003/.
@misc{reglegbrief_RLB_F_INT_UNTC_BBNJ_HIGH_SEAS_BIODIVERSITY_AGREEMENT_2023_Q003,
author = {RegLeg Specialist Panel},
title = {Finding#2 — MGR retroactivity default inverted by AI},
year = {2026},
publisher = {RegLegBrief AI Hallucination Research},
note = {Hallucination finding Citation ID: RLB-F-INT-UNTC-BBNJ-HIGH-SEAS-BIODIVERSITY-AGREEMENT-2023-Q003},
url = {https://reglegbrief.com/regulators/j1/int/untc/bbnj-high-seas-biodiversity-agreement-2023/sectors/law_firms/legal/finding/INT-UNTC-INT-001-BBNJ-HIGH-SEAS-BIODIVERSITY-AGREEMENT-2023-v1-003/}
}