AXIOM
Docs
🔍AnalysisJune 23, 2026·6 min read

AIS_manipulation_dark Overtook STS Three Weeks Ago. 26,246 Alerts. Zero Acknowledged.

The dominant behavioral motif in the maritime queue crossed over the week of June 1. 26,246 AIS_manipulation_dark alerts. Zero acknowledgments across 46,935 queued alerts.

🔮
Axiom Intelligence
Axiom Platform · June 23, 2026
0
ack count
20482
total sts
2026-06-21
data as of
45
window days
46935
total all motifs
1
avg confidence sts
0.809
avg confidence dark
2026-06-01
crossover week start
154
total fraudulent doc
55.9
ais dark pct of queue
10884
peak week dark alerts
26246
total ais manipulation dark
33
total ais manipulation spoof
TopicsAIS-MANIPULATION-DARKMOTIF-ALERTSAIS-OPACITYCOMPLIANCE-QUEUEBEHAVIORAL-MOTIF

The Setup

The motif alert queue has a new volume leader. AIS_manipulation_dark — vessels entering dark periods at frequencies and durations inconsistent with operational or regulatory necessity — generated 26,246 high-severity alerts in the 45 days through June 21. ship_to_ship_transfer generated 20,482. fraudulent_documentation generated 154. Across all 46,935 alerts, acknowledgment count: zero.

The Chain

The crossover happened the week of June 1. In that single week, AIS_manipulation_dark fired 8,791 alerts while ship_to_ship_transfer fired 5,639 — the first time dark-manipulation motifs outnumbered STS in the same weekly window. The week before (May 25), the ratio was reversed: 500 AIS_manipulation_dark alerts to 894 STS.

The volume timeline:

| Week | AIS_manipulation_dark | ship_to_ship_transfer | |---|---|---| | May 11 | 34 | 646 | | May 18 | — | 611 | | May 25 | 500 | 894 | | June 1 | 8,791 | 5,639 | | June 8 | 10,884 | 6,257 | | June 15 | 6,037 | 6,406 |

AIS_manipulation_dark went from near-zero to 10,884 weekly alerts in four weeks. STS ran relatively stable — between 5,639 and 6,406 — across the same period.

The total gap at 45 days: 26,246 versus 20,482. AIS_manipulation_dark leads by 5,764 alerts and holds 55.9% of the combined queue.

There is a third theoretically active motif — AIS_manipulation_spoof — that last fired on May 15 (33 total alerts). It hasn't appeared since.

The Implication

AIS_manipulation_dark runs at average confidence 0.809. The confidence range across weeks is 0.804–0.818 — narrow enough to be stable, not noise. ship_to_ship_transfer runs at 1.000. fraudulent_documentation runs at 1.000.

An analyst triaging the queue by confidence would place every STS alert and every fraudulent_documentation alert ahead of every AIS_manipulation_dark alert. That sequencing puts the dominant behavioral signal — 55.9% of the total queue — at the back of the processing order.

The gap: 0.191 confidence points below STS. At a 0.82 confidence floor, approximately 4,970 AIS_manipulation_dark alerts would be filtered out entirely. Zero STS or fraudulent_documentation alerts would fail that threshold.

The structural consequence: the queue processing problem (zero acknowledgments across 46,935 alerts) matters less if confidence-ranked triage is the operating model. Under that model, AIS_manipulation_dark can't be addressed before STS and fraudulent_documentation even if analyst capacity existed. The motif that grew 300× in four weeks is also the one with lowest processing priority.

What to Watch

Whether the June 15 dip (6,037 alerts — the first week STS recaptured the weekly lead since May 25) continues downward or whether AIS_manipulation_dark re-asserts. If the motif volume is normalizing, the cumulative 45-day gap closes over subsequent weeks. If it levels around 6,000 per week, the two motifs approach rough parity, with STS leading on confidence.

The zero-acknowledgment total is the cleaner metric to track over time. Weekly composition can shift without any analyst touching the queue.

Limitations

The confidence score of 0.809 is a classifier output, not a validated accuracy rate. A motif that grows from 34 to 10,884 weekly alerts in four weeks could reflect a genuine behavioral change in the tracked fleet, a sensitivity adjustment in the underlying detection model, or a definition reclassification that pulled previously untagged events into this category. The absence of adjudication data makes these explanations indistinguishable. The 300× weekly growth rate is unusually steep for a genuine behavioral shift; whether this represents real signal or classifier drift should be the first question an acknowledgment workflow answers.


Data as of June 21, 2026. Source: motif_alerts table, Axiom Overwatch. 45-day observation window.