Over 85% of maritime AI disruptions arise from dirty data, highlighting the need for improved data integrity and cybersecurity measures.
New reports shared today by the maritime technology analysis group Thetius and the cybersecurity organization Cydome contain findings that will change the industry's perspective on artificial intelligence. According to the analyses, over 85% of the disruptions experienced in ship autonomy and route optimization stem from low-quality inputs referred to as 'dirty data.'
Key Findings: Why is Data Integrity Important?
Collecting data in a maritime environment is a much more challenging process compared to terrestrial systems. The prominent 'data integrity' issues highlighted in the analyses are as follows:
Sensor Deviations: Sensors that become miscalibrated over time due to saltwater, corrosion, and vibrations send incorrect depth, fuel, or speed data to the AI model.
AIS and GPS Manipulation: Increased GPS jamming and spoofing activities, especially in conflict zones, trigger the input of 'corrupted data,' leading AI systems to make erroneous decisions.
Data Silos: The use of incompatible data formats among different systems on the ship (engine room, bridge, loading) prevents the AI from conducting a holistic analysis.
New Front in Cybersecurity: Data Poisoning
Another critical point highlighted in the report is cybersecurity risks. It is noted that cyber attackers are no longer just attempting to lock down the system; instead, they are resorting to the method of 'data poisoning.'
'Attackers are now ensuring that the AI makes incorrect but 'seemingly reliable' decisions by altering the data that feeds the AI in very small proportions, rather than stopping the system. This has become the hardest cyber risk to detect.' — Cydome Security Report 2026.
What Should the Industry Do?
Analysts emphasize that maritime companies should focus on three areas instead of 'smarter algorithms':
Data Governance: Ensuring the accuracy of data from the moment it is sourced until it is transformed into a decision.
Cross-Verification: Ensuring that the AI relies on data supported by independent sources such as radar and cameras, rather than just a single data source (for example, only AIS).
AI Literacy: Training the crew to a level where they can recognize inconsistencies in the data, rather than accepting the AI's decision as 'absolutely correct.'
Source: SeaNews Türkiye
