Guyana's MV Barima Disaster Exposed a Manifest Gap AI Could Have Closed

The MV Barima left Georgetown on the evening of 18 July bound for Port Kaituma, a community in Guyana's Barima-Waini region reached mainly by water or expensive charter flights. It capsized overnight off the Essequibo Coast, between Waini in the west and Adventure in the east. Search teams have since widened the search area by 32 square miles. The government's working figure, based on ticket sales and boarding counts, is about 179 passengers and crew.

The death toll has been hard to pin down. Kaieteur News reported as many as 103 dead on 22 July, while other outlets and later government briefings gave lower totals as identification caught up with recovery. On Friday 24 July, Prime Minister Phillips gave the clearest official figure so far: 73 bodies recovered, 69 identified, and 30 people unaccounted for. The rescue count was revised to 76 on 21 July, after reports from different sources were checked for duplicates. Al Jazeera, Demerara Waves and Guyana Times have tracked the count since the sinking.

Those figures, 73 recovered, 76 rescued and 30 unaccounted for, add up to the working total of 179. That total is itself an estimate built after the sinking from ticket sales and boarding counts, because the manifest listed only 116 passengers and 17 crew.

32 of the first 67 survivors were not on the manifest

Guyana's initial probe found that at least 32 of the first 67 people pulled from the water alive, nearly half, were not on the Barima's passenger list. The rescue count later rose to 76. On the night of the sinking, rescuers and the people identifying the dead were working from a list that missed a large share of those aboard, and nobody knows how many of the missing and dead were also unlisted.

MARAD, Guyana's Maritime Administration Department, has avoided the word "overloaded". Director General Captain Stephen Thomas said the Barima was within its rated capacity of 390 passengers and 126 tonnes of cargo, and that the investigation would focus on whether weight was properly distributed. A ship can be under its weight limit and still be unstable if cargo and people were loaded in the wrong places without anyone recording where. Officials say they cannot yet account for the full cargo, and all staff involved in loading the vessel have been suspended.

The Barima's captain and at least one crew member were detained after testing positive for cannabis and remain in custody. President Irfaan Ali's cabinet has ordered an independent commission of inquiry and is reportedly considering involving the parliamentary opposition to protect the integrity of the investigation.

The Caribbean has seen this before

The MV Neptune capsized off Miragoâne, Haiti, in February 1993, on its way from Jérémie to Port-au-Prince. It carried no lifeboats or life vests, and an estimated 1,500 people drowned when it met a squall carrying far more passengers than it was built for. Overloaded boats have kept sinking along Haiti's coast and in the Mona Passage since, including a migrant boat that capsized off the Dominican Republic in 2012 with at least 41 dead, and one off Turks and Caicos that killed 17.

In each case nobody on shore had a verified count of who was aboard when the vessel left. Where a boat is the only practical link between remote communities and everywhere else, operators face constant pressure to take one more passenger or one more sack of cargo, because turning people away can strand them for days. The rules exist on paper, but at a dock with no cameras and no scale they are rarely enforced.

How boarding verification works

Airlines reconcile boarding counts against ticketed passengers before the doors close. A hinterland ferry can do the same with modest equipment:

• Camera counts at the gangway. Computer vision, software that reads camera images, counts people as they cross, independent of the clerk's clipboard.
• Load cells on the gangway or deck. Cargo and passengers are weighed as they board, and the system keeps a running total.
• ID or ticket scans on a phone. Each boarding passenger is matched to a ticket and a manifest entry that syncs to shore and cannot be edited afterwards.
• Departure clearance from shore. A dispatcher, not the operator under schedule pressure, sees the count, the weight and the manifest and clears the sailing only when they match.

With this in place, the failure MARAD is now investigating would have shown up before the ferry left the dock. The technology is in commercial use: Lloyd's Register puts the global maritime AI market at US$4.13 billion in 2024, growing about 23 percent a year, with 420 organisations adopting AI-based maritime tools that year, up from 276 the year before.

A camera does nothing if the person watching the feed is told to look away. What it removes is deniability. A mismatch that is logged automatically, and that a regulator can audit without asking the operator, becomes an enforcement decision someone has to own.

Tracking the vessel after it leaves

AIS transponders, the radio beacons ships carry, broadcast a ship's position, speed and heading. Models trained on normal vessel behaviour can flag a sudden stop, a route deviation or a vessel sitting lower in the water than its manifest predicts. Research on AIS anomaly detection, including transformer models that flag unsafe speed, route deviation and loitering, has matured quickly over the past two years. Such a system would not have stopped whatever capsized the Barima, but it would have raised the alarm within minutes, instead of leaving it to whoever first noticed the ferry was overdue.

The same tools speed up search and recovery. Drift models based on currents and wind narrow the search box, and computer vision on drone and satellite imagery can flag debris and people in the water without a person scanning every frame. For families waiting on the 30 still missing, that can mean days instead of weeks.

Cost and enforcement are the real barriers

MARAD and most ferry operators in the region, from Suriname's river boats to the small services linking the Grenadines, do not have a global shipping line's budget. A system built from a camera, a load cell and a phone-based ticket scan is now cheap enough that the obstacles are a procurement decision and a regulation requiring it.

The Caribbean AI Risk Management Council makes the same argument about ferries that it makes about a bank's credit model: build the safety controls before the public failure. Every Caribbean government already has free access to TurtleBird, the AI safety toolkit built through Maestro AI Labs, for the same reason. A shared regional standard for ferry and boat safety data, adopted by Guyana, Suriname, Haiti and the smaller inter-island operators, would let St Vincent or Dominica use the same tools a larger country could build alone.

MARAD's Captain Thomas says the Barima was not overloaded by the numbers on paper. Because 32 survivors were missing from the manifest, nobody, MARAD included, can say for certain whether that is true.

What to do next

1. MARAD and Guyana's Ministry of Public Works: before the inquiry reports, require a shore-side head count and a signed cargo weight for every government ferry departure to the hinterland, checked by someone other than the crew.
2. The commission of inquiry: publish a reconciled list of everyone aboard, and explain how the 179 working figure was reached when the manifest listed 133, so families and future planners know what the record-keeping failure was.
3. Regional maritime administrations: pilot camera counting, load cells and ticket scanning on one busy route for six months, and publish how often the live count differed from the paper manifest.
4. Ferry operators: introduce random drug and alcohol testing for captains and crew before departure, with results logged in the same system as the manifest.
5. CARICOM and the Council: agree a common data format for ferry manifests and AIS alerts, so small states can buy the same tools instead of each building its own.