AI Hallucinations in Maritime QHSE: Detection and Verification Toolkit

AI does not know what it does not know.

When an AI produces an incorrect answer, it does so with exactly the same tone, formatting and apparent authority as a correct one. A hallucinated MARPOL Annex reference looks identical to an accurate one. A fabricated circular number reads no differently from a real one.

This toolkit addresses that one specific failure mode — and gives you a verification process you can evidence.

What’s inside

• 10 warning signs your AI is hallucinating, each with a maritime example and the action to take
• 27 high-risk scenarios across ISM/SMS, risk assessment, incident investigation, regulatory compliance, audits, training and emergency preparedness — each with a detection question
• 5-step verification framework plus a pre-issue checklist with sign-off blocks
• 15 detection prompts written for maritime QHSE contexts
• Safe AI Use Policy template — 10 sections, ready to adapt into your SMS
• 6 human review checklists by task type
• AI verification record and running log — your evidence trail
• Print-ready quick-reference card

Source-verified edition. Every regulatory reference is listed in Appendix A with its status and primary instrument. Illustrative examples are labelled as illustrative. Corrections from the previous edition are published openly in the change log — because a toolkit about accuracy should be able to demonstrate its own.

Formats: 28-page PDF + editable Microsoft Word file containing the policy, checklists, records and prompt library.

Licence: Single-organisation internal use across your managed fleet and offices. Not for redistribution or resale. Multi-fleet and group licences are available through sales@axiomordo.com.

General guidance, not legal or regulatory advice. Confirm all references against the current primary instrument for your vessel, flag and trade.