How to Reduce False Positives in Vessel Screening Without Missing Real Risk
What’s inside?
At a Glance
- Turning down a maritime screening system’s sensitivity reduces false alerts and real ones together, so the queue gets quieter while the exposure gets larger.
- Most false positives trace to a small number of data failures, including GPS jamming-generated port calls, AIS gaps misread as deliberate dark activity, and identity confusion after a flag or MMSI change.
- At one energy major, high false positive rates drove self-rejection of 46 to 50% of vessels and a recurring three-day manual exercise to pre-approve around 5,000 of them.
- Removing GPS jamming-generated activity for a single customer over one year lowered the risk score on roughly 495 vessels, with 260 moving from High to Low.
- The regulatory question is not whether you screened a vessel, it is how you determined it was not a risk.
Screening Alerts Are a Volume Problem Before They Are Anything Else
Every compliance team that touches maritime trade runs vessels through a screening system before a transaction proceeds. The system checks the vessel against sanctions lists, looks at its recent movements, flags anything that resembles a risk indicator, and produces an alert.
An alert is a referral, not a finding. It says a human should look at this. Someone then investigates, decides whether the concern is real, and records the outcome.
That process works when the alerts are mostly real. It breaks when they are not.
A false positive is an alert that fires on a vessel doing nothing wrong. Every one of them consumes analyst time that produces no risk finding, and in a high-volume book they arrive faster than a team can clear them – which spawns alert fatigue. And the usual response is to adjust the system’s sensitivity threshold, meaning the level of suspicion required before an alert fires.
Raise the threshold and fewer alerts fire. Fewer false positives, and also fewer true ones.
That trade-off is real, and it is the wrong trade-off, because it treats a precision problem as a volume problem. A screening system producing false positives at scale cannot reliably distinguish a vessel that looks like a risk from a vessel that is one. Moving the threshold changes how many of each you see. It does not improve the system’s ability to tell them apart.
Telling them apart requires context. Not only what the AIS record shows, but what the vessel has done over time, who owns it through how many layers, how its behavior compares to vessels confirmed as illicit, and whether the data source that triggered the alert is even reliable in the environment where the event occurred.
A sensitivity setting supplies none of that.
The Noise Has Specific Causes
False positives in vessel screening are not randomly distributed across the fleet. They cluster around four specific failures in the data the screening system reads, and each one produces a recognizable kind of bad alert.
The first is data that describes events that never happened.
The second is missing data read as deliberate concealment.
The third is a vessel’s identity changing faster than the screening record.
The fourth is correct behavior that looks wrong without context.
Each needs a different answer, and none of them is a threshold. The first failure is the least intuitive, because the data is not missing. It is present, detailed, and describes something that did not occur.
GPS Jamming Generates Activity That Never Happened
In a GPS-jammed area, injected coordinates place vessels at positions they never occupied. The system records a port call, a meeting, or a proximity event, and the screening engine treats it as real.
Across a single week of concentrated jamming in the Middle East Gulf, Windward removed 2,056 port calls across 845 vessels and 20,848 meetings across 695 vessels, none of which occurred.
The second failure is the opposite. The data is absent, and the screening system reads the absence as intent.
AIS Gaps Get Read As Deliberate Dark Activity
A vessel loses signal in a congested port and the alert fires as though it went dark on purpose. In the Middle East Gulf, switching off AIS has become normal practice rather than a red flag.
Windward recorded dark convoy passages accounting for 47% of Strait of Hormuz transits in August 2026 and 52% in September, as vessels crossed with AIS off for crew and vessel safety. A screening model treating every gap in that corridor as concealment would flag the majority of legitimate traffic.
Compliance teams tell us this is where much of their Iran-related false positive volume comes from. Vessels switch off AIS for crew and vessel safety rather than for concealment.
The third failure is not about the data being wrong. It is about the screening system matching against the wrong vessel.
Identity Changes Break the Match
A vessel that changes flag, name, or MMSI can trigger screening against its former registration rather than its current one, or be confused with a different vessel carrying a similar history.
The fourth failure involves no data error at all. The record is accurate, and the alert is still wrong.
Legitimate Transits Look Like Approaches
A vessel routed correctly past a sanctioned port appears suspicious without route context. The alert describes proximity. It does not describe intent, destination, or whether the vessel ever stopped.
None of these four is a tuning problem. Each comes from what the screening system is reading, or failing to read, which means each has an answer in the data rather than in the sensitivity setting.
Context Resolves What a Trigger Cannot
What separates a real alert from a false one is almost always something outside the event that triggered it. The vessel’s own past behavior, the behavior of comparable vessels, and whether the physical facts support what the record claims.
A screening system that evaluates each event in isolation cannot reach any of that. One that compares the event against the vessel’s history can.
Take the GPS jamming case. A port call recorded in an area with active jamming during the arrival hour is suspect. A port call recorded in that same area by a vessel that has made clean, unjammed visits to that port in the preceding six months is probably real.
One clean historical visit is enough to validate the current call. That single comparison separates a manufactured event from a genuine one without any analyst touching it.
The same logic applies across the other categories. A speed and distance feasibility check establishes whether an alleged port call was physically possible given the vessel’s known positions before and after.
Ownership network analysis surfaces whether a beneficial owner shows a pattern of association with sanctioned entities across a fleet rather than on the single hull being screened. Pattern-of-life analysis built on multi-year behavior distinguishes a vessel making an unusual call from a vessel behaving in a way that is structurally consistent with illicit trade.
None of these reduce sensitivity. They replace a binary trigger with a calibrated assessment, which is a different thing entirely.
The Cost of Getting This Wrong Is Measurable
The effect shows up in two directions.
One major energy company reported to Windward, self-rejecting 46 to 50% of vessels under evaluation, losing business to competitors as a result, and running a recurring three-day manual exercise every two weeks to pre-approve a list of roughly 5,000 vessels using several data sources in parallel. Its compliance team described difficulty justifying those rejections to senior management without better data behind them.
That is the business cost. The compliance cost runs the other way.
For one customer over a twelve-month period, removing GPS jamming-generated activity lowered the risk score on approximately 495 vessels. Around 260 moved from High to Low, 160 from High to Moderate, and 75 from Moderate to Low. Those were not borderline calls being softened. They were vessels carrying elevated risk scores built on events that had not occurred.
A compliance team working that book is not just seeing too many alerts. It is prioritizing against a risk ranking that is wrong.
Clearing an Alert Is a Documentation Problem
There is a regulatory argument underneath the efficiency one, and for compliance leads it is the more important of the two.
When a team clears a vessel, the question a regulator asks later is not whether it was screened. It is how the determination was reached that it was not a risk.
“AIS showed it was not in a sanctioned port” is a thin answer, particularly if the AIS record in question came from a jamming-affected corridor. A multi-sensor assessment is a materially stronger one.
The vessel’s AIS showed proximity to the terminal, satellite imagery showed it anchored offshore with no draft change consistent with discharge, its ownership chain resolved clean across seven layers, and its behavior over eighteen months shows no prior association with sanctioned terminals.
That resolves in minutes rather than hours, and more importantly, it leaves a record of why the decision was made rather than only that it was.
One further point worth stating. Not every alert that clears was a false positive. A genuine designated-vessel encounter can be cleared on the strength of a specific license and supporting documentation, and that is a correct finding, not a system error. The distinction matters for how a team measures itself. Counting cleared alerts as noise will push a program toward exactly the threshold-raising it should be avoiding.
Fewer Alerts Is the Byproduct, Not the Goal
The objective is not a quieter queue. A quieter queue is achievable in an afternoon by moving a slider, and the cost arrives later.
The objective is that each alert reaching an analyst carries enough context to be resolved with confidence, and that each clearance leaves a record capable of surviving external review. Volume falls out of that as a consequence.
A screening system that fires on what a vessel appears to have done will keep producing noise, because appearance is what deceptive operators control. A system that assesses what a vessel has actually done, across time and across sources, produces fewer alerts and better ones.
Written by Maya Romi, Windward Maritime Intelligence Senior Content Manager.
Trending
- The EU’s 18th Sanctions Package Lookback Started. Trading Russian Products? You're At Risk. Nov 24, 2025
- Tanker Freight Rates Hit Five-Year High Amid Russian Oil Sanctions Shake-Out Nov 6, 2025
- Sanctioned, Stateless, and Still Sailing: Expert Insights from the Frontlines of Maritime Sanctions Nov 3, 2025