Marine companies in the Gulf have more to read each morning than at any point in recent memory, and AI agents are good at reading. The same companies are rightly unwilling to let an AI tool write into the records their operations run on. This article describes a way of using an agent that keeps both positions intact: the agent does the reading and proposes, and a person decides what becomes part of the record.
The figures below come from UKMTO, the Joint Maritime Information Center and the ICC International Maritime Bureau, checked on 21 November 2026.
How Much There Is to Read
| Figure | Period | Source |
|---|---|---|
| 101 warnings and 13 advisories | Year to 28 August 2026 | UKMTO Voluntary Reporting Area Overview |
| 80 updates to one advisory note | 1 March to 4 August 2026 | Joint Maritime Information Center |
| 51 piracy and armed robbery incidents | January to September 2026 | ICC International Maritime Bureau |
The last of those is good news: it is the lowest nine-month figure the Bureau has recorded since 1992, although its report also notes a rise in Somali piracy. The first two show the other side. An advisory note that is updated roughly every two days for five months is a document somebody has to read roughly every two days, on top of port notices, insurers' circulars and the maritime press.
Why Teams Keep AI Away From Their Records
The caution is well founded. A language model can summarise a warning accurately ninety-nine times and then state something confidently that the source never said. In an operations room, a wrong entry against a vessel or a route is worse than no entry, because people act on what the system shows them.
The usual response is to keep AI tools out altogether, which leaves the reading to be done by hand. A better response is to design the system so that the agent's mistakes cannot reach the record without a person seeing them first.
Four Rules for an Agent People Will Trust
It drafts and never decides
Everything the agent produces is a suggestion in a queue. A person accepts, edits or rejects each item, and only accepted items go any further.
Every item carries its source
Each suggestion shows where it came from and links back to it, so the reviewer reads the claim against the original and not on trust.
People set the ranks
Each source has a credibility rank chosen by the team. The rank decides how an item is handled, and it moves with what the team accepts and rejects.
It writes to its own database
The agent cannot touch operational records. It writes to a separate store, and accepted items are copied across by a person's action.
Ranking the Sources
An official warning and an unconfirmed post should not arrive looking the same. A simple set of ranks, set by the people who know the sources, decides how far each item can travel.
| Rank | Typical sources | How an item is handled |
|---|---|---|
| Official | UKMTO, the IMB Piracy Reporting Centre, port authorities | Accepted after a single review |
| Trusted industry | Insurers' circulars, established maritime press | Reviewed, and usually accepted |
| Open web | News sites, social posts | Held for review, with the source shown |
| Blocked | Sources the team has ruled out | Never reaches the queue |
Ranks should not be fixed for ever. A source the team keeps accepting has earned a higher rank, and one it keeps rejecting has earned a lower one, so the queue improves with use.
Two Databases, One Direction
The separation that makes the rest safe is structural. The agent and the reviewers write to an intelligence database. The operational database, which holds vessels, voyages, jobs and clients, only reads from it. When a person accepts an item, their action copies it across with its source attached.
This means there is no path by which an agent can change an operational record, whether through a fault, a bad instruction hidden in a web page it has read, or a simple error. The worst an agent can do is put a poor suggestion in a queue, where a person will reject it.
Your own people outrank every feed
What a master says on the phone, what a team on board sees and what the office already knows are usually better than anything an agent can find. A good system lets that be logged against the vessel in seconds, with who gave it and when, and ranks it above third-party sources.
What the Team Gains
The duty officer starts the watch with a short queue in place of a long list of sites. The risk lead controls which sources are trusted and has a record of every decision taken on what they reported. Client-facing staff find accepted items already attached to the vessels and routes they affect. And what one person learned on one shift is still there for the next.
The same pattern applies outside maritime work, wherever an organisation wants an agent to gather information that people then act on. We have written separately about the UAE's agentic AI mandate and the readiness gap.
Common Questions
Yes. Both publish their reports openly, and an agent can read them each day along with any other sources a team chooses, then queue what it finds for review.
By never treating its output as fact. Each item carries its source and a link, a person checks it against that source, and nothing enters the record until it has been accepted.
So that it has no means of altering operational records. The operational system reads from the intelligence store, and only a person's action moves an item across.
The organisation using it, on its own account with the model provider and with a spending cap it sets, so the running cost is visible and under its control.
No. Reports from a service you already subscribe to can be stored against the voyages they relate to, alongside what the agent finds and what your own people report.
Sources
- UKMTO Voluntary Reporting Area Overview, Report No. 4 (28 August 2026)
- Joint Maritime Information Center: Update 080 to JMIC Advisory Note (4 August 2026)
- ICC International Maritime Bureau: Maritime piracy incidents remain at historic lows despite Somali piracy concerns (15 October 2026)
BY BANKS designs and builds bespoke platforms for marine companies in the UAE, including maritime intelligence software built on the pattern described here. Our wider work with AI is on the AI and automation page, and if you are weighing this up, get in touch.
Maritime security reporting changes daily. Treat the linked sources as the reference and this article as a summary of them. Nothing here is security or navigational advice. BY BANKS is an independent software engineering company. References to UKMTO, the Joint Maritime Information Center, the ICC International Maritime Bureau and any other named bodies or products are descriptive only and do not imply endorsement, partnership or affiliation.
Ready to Build Something?
If this resonated, let's talk about how we can apply these ideas to your business.
Start a Conversation