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

The MV Barima left Georgetown on the evening of July 18 bound for Port Kaituma, a remote community in Guyana's Barima-Waini region reachable mainly by water or costly charter flights. It never arrived. The vessel capsized overnight off the Essequibo Coast, in the stretch of ocean between Waini in the west and Adventure in the east, an area search and rescue teams have since expanded by 32 square miles as recovery work continues. The government's working figure, based on ticket sales and boarding counts, is that around 179 passengers and crew were aboard.

The death toll has been genuinely hard to pin down, and that in itself says something about the state of the record-keeping. Kaieteur News reported figures as high as 103 confirmed dead on July 22, while other outlets and later government briefings gave lower running totals as identification caught up with recovery. By Friday, July 24, Prime Minister Phillips gave the clearest official figure to date: 73 bodies recovered, 69 of them positively identified, and 30 people still unaccounted for against the working passenger count. Al Jazeera and regional outlets including Demerara Waves and Guyana Times have tracked the toll rising almost daily since the sinking. Families in Georgetown and Port Kaituma are still waiting on confirmation for people who have not been found.

Nearly Half the Survivors Weren't on the Manifest

The detail that should worry every ferry operator in the region is not the weather or the age of the vessel. It is the manifest. Guyana's initial probe found that at least 32 of the 67 people pulled from the water alive were never recorded on the Barima's passenger list. That is not a rounding error in a paperwork process. It means that on the night the ferry sank, the people responsible for finding survivors and identifying the dead were working from a document that undercounted who was actually on board by close to half.

MARAD, Guyana's Maritime Administration Department, has been careful about the word "overloaded." Director General Captain Stephen Thomas stated that the Barima was within its rated capacity of 390 passengers and 126 tonnes of cargo, and that the investigation would instead focus on whether weight was properly distributed across the vessel. That distinction matters less than it sounds. A ship can be under its total weight limit and still be dangerously imbalanced if cargo and unlisted passengers were loaded in the wrong places, at the wrong time, without anyone recording where. The government's own account acknowledges this: officials say they cannot yet account for the Barima's full cargo, and all staff involved in loading and managing the vessel have been suspended while investigators determine whether people and goods were boarded without being manifested at all.

The human factor compounds the paperwork one. The Barima's captain and at least one crew member were detained after testing positive for cannabis, and remain in custody. President Ali's cabinet has ordered an independent commission of inquiry, reportedly weighing whether to involve the parliamentary opposition, specifically to protect the investigation's integrity and prevent evidence from being altered before it reports.

The Caribbean Has Been Here Before

Guyana's tragedy sits inside a pattern the region has lived through repeatedly, at far greater scale. The MV Neptune capsized off Miragoâne, Haiti, in February 1993, en route from Jeremie to Port-au-Prince. It carried no lifeboats and no life vests, and an estimated 1,500 people drowned when it ran into a squall while carrying far more passengers than it was built for. Overloaded vessels have kept sinking along Haiti's coast and across the Mona Passage between Hispaniola and Puerto Rico in the decades 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.

The common thread across every one of those disasters, and across the Barima, is not bad luck or even bad weather. It is that nobody on shore had a verified, real-time count of who was actually on the vessel when it left. In places where ferries and boats are the only practical connection between remote coastal or hinterland communities and everywhere else, that gap becomes structural. Operators face constant pressure to carry one more passenger, one more sack of cargo, because turning people away means stranding them for days. Regulation exists on paper. Enforcement, in the moment a boat is loading at a dock with no cameras and no scale, often does not.

The Manifest Problem Is Solvable, and It Doesn't Need New Physics

This is where the technology conversation actually gets useful, because the fix here is not exotic. Passenger and cargo verification at the point of boarding is a solved problem in industries that decided it mattered enough to build. Airlines reconcile boarding counts against ticketed passengers before a single door closes. What the Barima's loading process lacked was the equivalent for a hinterland ferry: a system that counts who and what actually crosses the gangway and checks that count against a manifest nobody can quietly edit afterward.

Computer vision cameras mounted at a vessel's boarding point can count people as they cross, independent of whatever a clerk writes on a clipboard. Load cells built into the gangway or deck can weigh cargo and passengers as they board, giving a running total that a dispatcher, not just an operator with an incentive to sail on schedule, can see before granting departure clearance. Reconcile that live count against ticket sales and a cloud-synced manifest, and the exact failure MARAD is now investigating, people and cargo boarding without ever being recorded, becomes visible before the ferry leaves the dock, not after it sinks. According to Lloyd's Register, the global maritime AI market reached $4.13 billion in 2024, growing at a 23 percent annual rate, with 420 organisations adopting AI-based maritime tools that year, up from 276 the year before. This is not speculative technology. It is being deployed, at scale, in fleets and ports outside the Caribbean right now.

None of that fixes a corrupt or careless loading process by itself. A camera that counts unlisted passengers boarding does nothing if the person watching the feed is told to look away. What it does is remove plausible deniability. Right now, when 32 of 67 survivors turn out to be off the manifest, the honest answer to "how did that happen" is that nobody was actually tracking it in real time. A verification system that flags the mismatch the moment it happens, and that a regulator can audit independently of the operator, turns a systemic gap into an enforcement decision. That is a meaningfully different problem, and a more solvable one.

Track the Ship, Not Just the Dock

The second half of this is what happens once a vessel is already at sea. Modern AIS tracking systems, the transponders that broadcast a ship's position, speed, and heading, generate a continuous stream of data that AI models trained on historical vessel behaviour can monitor for anomalies: a sudden change in speed, an unexplained deviation from the expected route, or a draft reading that suggests a vessel sitting lower in the water than its cargo manifest would predict. Research on AIS-based anomaly detection, including transformer models built specifically to flag unsafe speed, route deviation, and loitering, has matured well beyond the research-paper stage over the past two years. A system built on that model would not have prevented whatever combination of overloading, weather, or human error capsized the Barima. It would have flagged the vessel's distress within minutes rather than leaving the alarm to be raised by whoever first noticed it was overdue.

The same class of tools speeds up what comes after. Search and rescue teams expanding a recovery zone by 32 square miles are, in effect, running a grid search across open water largely by eye and by drift calculation. AI-assisted current and drift modelling, combined with computer vision applied to drone and satellite imagery to classify debris and objects in the water automatically rather than requiring a human to scan every frame, cuts the time between "we don't know where to look" and "we have a search area." For families waiting on identification of the 30 people still unaccounted for as of July 24, that is not an abstract efficiency gain. It is the difference between days and weeks.

What Stands Between This and a Guyanese Ferry Terminal

The honest constraint is cost and enforcement, not invention. Guyana's Maritime Administration Department, like most hinterland and inter-island ferry operators across the Caribbean, from Suriname's river transport network to the small boat services connecting the Grenadines, does not run on the budget of a global shipping line. A boarding-verification system built from a camera, a load cell, and a phone-based ID scan is now cheap enough that the barrier is a procurement decision and a regulatory mandate, not a research problem. That is a genuinely different situation from ten years ago, when this kind of monitoring required infrastructure only a large port could afford.

I chair the Caribbean AI Risk Management Council, and the argument the Council makes about AI risk applies here just as directly as it does to a bank's credit model or a government's chatbot: the region cannot treat safety infrastructure as something to add once a system has already failed publicly. Every Caribbean government already has free access to TurtleBird, the AI safety toolkit built through Maestro AI Labs, because the alternative, building this kind of tooling from scratch after a disaster, is both slower and more expensive than building it before one. A shared regional standard for ferry and boat safety data, built once and adopted across Guyana, Suriname, Haiti, and the smaller inter-island operators, would let a country the size of St. Vincent or Dominica adopt the same verification tooling a larger economy could build alone, at a fraction of the cost, the same logic that let a three-person team in Kingston build a sovereign AI model this year instead of paying the usual multi-million dollar price tag to do it from scratch.

MARAD's Captain Thomas says the Barima wasn't overloaded by the numbers on paper. The 32 people missing from its manifest are the reason nobody, including MARAD, can say with certainty whether that is true.