Deception Check ← all research

Research method and scope

Deception Check | September 16, 2026

What we counted

The blog uses retained records from our S7-emulating and DICOM-emulating internet-facing research decoys. The shared window starts September 14, 2026 at 21:00 UTC and ends September 15 at 21:00 UTC, with the endpoint excluded.

These operations describe device identification and connection negotiation; they do not establish PLC changes, patient-data access or confirmed attacks. The observations come from an external research fleet, separate from CISA's internal-deployment guidance.

Source context

Known operator and fleet addresses were excluded. Other public source addresses were compared with a saved August 25 research-scanner network list: 47 of the 49 S7 reads and all five DICOM requests matched. A match identifies a listed network, not a verified operator or intent. Unmatched traffic is not automatically malicious.

Verification and limits

The review checked event-file hashes against backup manifests, parsed the records, removed repeated event identifiers within each node, and recounted the operations. The main window falls within every selected node's retained timestamp range.

Retention gaps between backups and reported logging-admission failures mean this is not a lossless or continuous fleet census. Counts are of retained operations, not unique attackers or customer outcomes. These native-protocol responses were recorded as deterministic/scripted; this analysis does not measure adaptive-model performance or customer detection-time improvement.

The practical use of this example is to show why an alert benefits from a record of the requested operation and its source context.

Back to the guide breakdown