Industries / Transport and vehicles / Maritime and shipping / Ports and terminals
Safety
Port safety covers the people who work among ships, cranes, stacked containers and moving machines, and the rules that keep cargo from crushing, striking or burying them.
4
Milestones achieved, from 1929 to 2024.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1929
Dockers get a treaty
The International Labour Organization (ILO), the UN agency for work, adopted a convention on protecting dockers against accidents in 1929. It was revised in 1932 as Convention No. 32, and the ILO's first code of practice for dock work followed in 1958.
Credit
ILO
1979
Port safety rules rewritten
In 1979 the ILO replaced Convention No. 32 with the Occupational Safety and Health (Dock Work) Convention (No. 152) and Recommendation No. 160. They cover all work in ports where goods or passengers are loaded onto or unloaded from ships. The ILO says a number of countries have ratified the convention and many others have used it as the basis for port-work laws, and its Governing Body has decided it is still up to date.
Credit
ILO
2016
People and vehicles kept apart
A revised ILO code of practice on safety and health in ports was adopted in Geneva in November 2016 by a meeting of 24 experts named by governments, employers and workers. It says vehicles and mobile plant are now among the main causes of fatal and serious accidents in ports, and that people should be separated from vehicles whenever practicable. It replaces the 2005 edition and is guidance, not law.
Credit
ILO Meeting of Experts
2024
A national code after deaths
On 19 April 2022 a stevedore at the Port of Auckland, New Zealand, was crushed when a suspended 40-foot container was lowered onto them. On 25 April 2022 a stevedore loading coal at Lyttelton was found dead under coal on a ship's deck. New Zealand's Transport Accident Investigation Commission found that the sector had no best-practice guidelines, no minimum training requirements and few ways to share safety information. An Approved Code of Practice for loading and unloading cargo, written by ports, unions and the regulator Maritime NZ, took effect on 29 November 2024; inspectors and courts can refer to it.
LessonA review after the deaths found safety standards differed from port to port.
Credit
Port Health and Safety Leadership Group; Maritime NZ
Watch for
A larger count of deaths
ICHCA International, a cargo-handling industry association, counted over 500 deaths of port workers, contractors and visitors from 2000 to June 2025 in its members-only Severe Risks Dashboard. Crush by cargo was the leading cause, at over a quarter of incidents. In July 2026 its chief executive, Richard Steele, wrote of more than 25 years of fatality data covering over 1,100 incidents and almost 1,700 casualties worldwide.
Achieved whenthe larger dataset is published with its sources, so that others can recount it.
Claimed by
ICHCA, July 2026
Open
01
An open register of deaths
A free public list of people killed in cargo handling at ports worldwide, with the date, port, task and cause of each death and a link to an official report or a news account.
The briefCredit · unclaimed02
Under-load alerts put to test
A public set of labeled video clips from quay and yard cranes, with every moment a person stands beneath a suspended container marked, and a scoring script that measures how many of those moments an alert system catches and how many false alarms it raises. It runs on recordings only and controls nothing.
The briefCredit · unclaimed
Open 01 · the brief
An open register of deaths
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A free public list of people killed in cargo handling at ports worldwide, with the date, port, task and cause of each death and a link to an official report or a news account.
Why it is still open
In 2018 Michael Cross of the Maritime Union of Australia told the International Transport Workers' Federation (ITF), "There is no central data point to keep such statistics." The nearest thing, ICHCA's dataset, is for members only, and ICHCA says it "is not exhaustive."
Built with what exists
A web agent that gathers investigation reports and news (see the Note "Deleting 80% to start over"); an AI model that answers fixed questions about each report, such as cause and task (see the Note "A thousand questions, one pass"); a public Git repository.
First step
Enter the two 2022 New Zealand deaths from the Commission's report, then the port deaths in the latest annual reports of three national accident investigators.
Done looks like
300 deaths from at least 15 countries, each with a primary source and coded into ICHCA's cause groups, with two people checking the coding on a random 50 and the agreement rate published.
Open 02 · the brief
Under-load alerts put to test
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of labeled video clips from quay and yard cranes, with every moment a person stands beneath a suspended container marked, and a scoring script that measures how many of those moments an alert system catches and how many false alarms it raises. It runs on recordings only and controls nothing.
Why it is still open
Camera systems that watch for people under suspended loads are on sale. viAct says its system detects a worker entering the area beneath a suspended load and reports an 82% fall in unsafe entries at one US equipment plant, by its own case study. ICHCA names crush by cargo as the leading cause of port deaths, and the 2022 Auckland death happened under a suspended container. No independent test at a container terminal, and no public labeled clips, turned up in a web search on 18 November 2026.
Built with what exists
An open object-detection and segmentation model such as YOLO as a baseline, run on a laptop with a graphics chip; a labeling tool such as CVAT.
First step
Ask one terminal for a day of recorded quay-crane video and label the under-load moments by hand.
Done looks like
200 labeled clips from at least two terminals, released with permission, and the share of under-load moments caught and the false alarms per hour of video for the open baseline and for at least one commercial system whose maker agrees to be tested.