864 Vessel Identity Reassignments Since April. Every One Flagged as Obfuscation. None Has Closed.

The Setup

Since April 5, 2026, the vessel identity graph has logged 864 mmsi_changed edges — a vessel's Maritime Mobile Service Identity swapped for a new one. All 864 sit on 864 distinct vessels: no vessel in this window has reassigned its MMSI more than once. Every one of the 864 carries obfuscation_flag = true. Every one is still open — zero carry a valid_to timestamp. Average confidence decay on the edge: 0.700. The most recent reassignment logged was August 5; as of this writing the graph has gone seven days without a new one.

The Chain

Compare that to renamed_to, the graph's other identity-mutation relation type: 1,258 edges recorded across the same history, the same 100% open rate — but zero flagged for obfuscation. Not one. The graph currently treats a hull's registered name change as background noise and its MMSI reassignment as adversarial by default, and it draws that line with no middle ground: 0% on one side, 100% on the other. There is no edge in either bucket sitting at, say, 40% obfuscation confidence waiting on a closer look — the classifier isn't uncertain about MMSI changes, it's decisive. But "decisive" and "adjudicated" are not the same claim. Nothing in temporal_edges records who reviewed any of these 864, when, or why all of them remain open four months after the earliest was logged.

For scale, this pattern isn't unique to MMSI changes: shell_hop edges (511 logged, covered in a prior analysis) show the same 100%-flagged, 0%-closed signature. What's new here is that MMSI reassignment runs the identical pattern at a larger n, on a relation type that hasn't been examined on its own before, and it's the more surgical of the two moves — a shell hop restructures commercial management; an MMSI change touches the transponder identity a vessel broadcasts on every AIS ping.

The Implication

An MMSI reassignment breaks the link between historical AIS track and current broadcast identity without touching the hull's IMO number, its name, or its ownership chain — it's a narrower, more targeted move than a full shell hop. If 100% of them get flagged and 0% ever get closed, one of two things is true: either MMSI reassignment genuinely has no legitimate use case anywhere in this tracked fleet population — plausible, but a strong enough claim that it deserves a documented basis somewhere in the pipeline — or the flag is a default state nobody has built a path out of. Either way, a sanctions-screening or counterparty-diligence process pulling from this edge type today is pulling 864 permanently-open allegations with no adjudicated outcome attached to any of them, and no way to distinguish a genuinely evasive reassignment from a re-registration that happens to coincide with an MMSI update.

What to Watch

Whether the seven-day silence since August 5 continues, or a new reassignment cluster begins — weekly volume has ranged from single digits to several dozen since May with no obvious pattern yet. Watch whether any of the 864 ever receives a valid_to timestamp; that would be the first evidence the graph has a resolution path at all, rather than a one-way accumulation. Watch too whether renamed_to's 0% flag rate holds as volume grows, or whether it's simply a relation type nobody has built an obfuscation heuristic for yet.

Limitations

This reads relation-level metadata — obfuscation_flag, valid_to, confidence_decay — off temporal_edges, not the underlying evidence for any individual reassignment. It says nothing about whether the flag is well-calibrated case by case, only that the graph currently draws no distinction among the 864. The sample is fleet-wide since April 5 with no filtering by flag state, vessel class, or region, so it may bundle genuinely different populations — legitimate re-registration and active evasion — under one relation type. n=864 is small relative to the graph's roughly 180,000 total temporal edges, and a 100% rate on a few hundred records is a narrower claim than the same rate would be at ten times the volume.

Data as of 2026-08-12. Source: temporal_edges (relation_type = 'mmsi_changed', 'renamed_to'), Axiom Overwatch temporal graph.