Preparing for SWIFT CSCF v2024 Control 2.9 anomaly detection attestation is rarely a tooling problem. Most institutions already log SWIFT activity. What they lack is a documented, defensible definition of normal correspondent traffic, and without it detection either floods the team with alerts or stays silent when it matters.
This article describes three baselines to establish, four thresholds to build on them, and how to prepare the evidence. The SWIFT Customer Security Controls Framework is set by SWIFT, and Saudi institutions also work within the SAMA Cyber Security Framework. Thresholds below are starting points to be calibrated to your own data, not fixed values.
When you need to do this work
Consider it now if any of these apply:
- Your attestation or independent assessment is within the next two quarters.
- Alert rules for SWIFT traffic were copied from a generic template and nobody can explain their values.
- Analysts routinely close SWIFT alerts as expected behaviour without a record of why.
- Payment operations and security operations have never agreed what normal looks like.
- Your correspondent network or transaction mix has changed since rules were last tuned.
The three baseline metrics
An assessor will want to see that each metric is measured, owned, reviewed and tied to an alert or review action.
1. Message-volume variance by correspondent relationship
Measure daily and weekly message counts and value for each correspondent, not for the institution as a whole. Aggregate volume hides a single relationship behaving oddly. Record the typical range, the seasonal pattern (month-end, salary cycles, Hajj and Ramadan periods) and the normal spread around it.
2. Off-hours transaction frequency
Define business hours per operating unit, including weekends and Saudi public holidays. Then record how many messages are normally released outside those hours, by user, role and message type. Some off-hours activity is legitimate, for example treasury cut-offs in other time zones. The baseline separates that from the unexplained.
3. Destination-country deviation
For each correspondent and business line, record which destination countries and currencies appear and in what proportions. A payment to a country never seen in a relationship's history carries more meaning than a large payment to a familiar one.
The four traffic-pattern thresholds
Build thresholds from the baselines above, and make each one a combination of signals rather than a single trigger. This is what keeps false positives manageable in high-volume environments.
- Volume excursion against the correspondent's own range. Alert when a relationship's volume or value moves beyond its own seasonal band, measured over a defined window, rather than against a global number. Calibrate the band from at least a full annual cycle where history exists.
- Off-hours release outside the approved pattern. Alert when messages are released outside defined hours by a user, role or message type that has no history of doing so. Approved exceptions should be listed and reviewed, not silently suppressed.
- First-seen destination. Alert on a destination country or currency new to that correspondent or business line, and raise severity when it coincides with a new beneficiary or a value above the relationship's norm.
- Combined deviation. Escalate when two or more of the above occur within a short period, or alongside a change in operator activity such as a new session pattern or a recent credential reset. Single weak signals go to a review queue; combinations go to immediate triage.
Tiering alerts this way matters more than the exact numbers. Weak signals reviewed daily and combined signals triaged at once give analysts a workable load and give assessors a clear rationale.
Avoiding alert fatigue
- Tune per relationship, not globally.
- Document every suppression with an owner, a reason and an expiry date.
- Review thresholds on a fixed schedule and after any change to the correspondent network, message types or operating hours.
- Track the proportion of alerts that lead to action, and use it to retire noisy rules.
- Rehearse the response: who is called, who can hold a payment and how quickly.
What a good outcome looks like
By attestation, you should be able to show:
- A written baseline for each of the three metrics, with its data source and review date.
- Thresholds that trace back to those baselines.
- Alert records showing triage decisions and rationale.
- Evidence that payment operations and security operations both agreed the definitions.
- A tested escalation path for suspected compromise.
How we run it
Our SWIFT CSCF Assessment & Attestation service is an independent assessment of your SWIFT environment against the current Customer Security Controls Framework. The work follows four stages: scope, assess, remediate and attest.
We start by confirming your architecture type and the in-scope components, because this determines which controls apply. We then assess each control with evidence, covering both mandatory and advisory controls, and Control 2.9 is examined on whether your baselines, thresholds and alert handling can be evidenced rather than merely described. Where gaps appear, we provide remediation support, and we then prepare the final report for your attestation.
The assessment phase takes 3–5 weeks. You receive an independent assessment report, a control-by-control evidence file and a gap remediation plan. The service aligns to SWIFT CSCF and SAMA CSF.
Next step
If your attestation is approaching and your anomaly-detection baselines are undocumented or untested, an early conversation will leave time to close gaps before the assessment window. Contact CISO Consulting to discuss SWIFT CSCF Assessment & Attestation for your institution.