RESPONSE NOTES / CVE-2026-106173
Who watches the SIEM
when the SIEM is the target?
The FortiSIEM file-write vulnerability is a reason to check more than patch status. Establish who can reach the service, investigate changes, and keep evidence outside the appliance you may need to investigate.
If the SIEM is the system under investigation, where will you look for evidence you can still trust?
A monitoring platform can be both a source of evidence and an incident target. Investigating it takes more than its own dashboard: establish exposure, investigate changes, and retain a record outside the appliance.
Start with exposure, not the dashboard
Identify the deployed version, the affected component's host, and the paths that actually reach it. “Internal” is an intended use, not an access control. A firewall rule, a routing boundary, and a verified allowed-peer list tell you more than the service's name.
Fortinet's interim guidance is to restrict TCP 27918. Preserve required FortiSIEM peer communication while removing unnecessary access; involve the platform owner before changing a production rule. Check relevant network segments through approved configuration review or safe reachability testing. Keep exploit and outage simulations in an isolated lab.
Version guidance: the primary sources disagree
The current Fortinet advisory lists 7.5.0 as affected and directs users to 7.5.1 or later. It lists 7.6 as unaffected and separately says FortiSIEM Cloud is unaffected.
However, Fortinet's CVE record also lists affected ranges in older branches:
| Branch | Range in the CVE record |
|---|---|
| 7.4 | 7.4.0 through 7.4.2 |
| 7.3 | 7.3.0 through 7.3.5 |
| 7.2 | 7.2.0 through 7.2.7 |
| 7.1 | 7.1.0 through 7.1.9 |
| 7.0 | 7.0.0 through 7.0.4 |
Its solution text also mentions FortiSIEMCloud 25.4.1 and FortiSOC 26.2.2. We have not reconciled that discrepancy with Fortinet. For an older branch or cloud deployment, ask Fortinet to confirm applicability and the supported upgrade path for your exact deployment. Do not turn an omitted row into an “unaffected” decision.
A short path from service access to execution
Horizon3 describes a gRPC service that does not require a client certificate. Crafted metadata permits traversal and filename manipulation, allowing a file overwrite. The published PoC targets a repeatedly executed script; its current source describes execution in the appliance's admin context, not root. Research analysis; PoC source.

Read the diagram as text
- Reach the serviceUnauthenticated access to the vulnerable gRPC service. Check: actual paths to TCP 27918.
- Write outside the intended pathTraversal and filename manipulation enable a file overwrite. Check: trusted file baseline and retained changes.
- Changed script runsThe published PoC uses a recurring script in the admin context. Check: execution and follow-on activity.
Recommended response, not an exploit stage: Retain useful evidence outside the appliance being investigated. Check: retrieval, access separation, and owner.
At each boundary, ask: what record would establish that this stage happened, and where would that record survive?
Keep three kinds of evidence distinct
01 / CONNECTION
Who reached the service?
External firewall and flow records can identify source, destination, time, and connection outcome. A non-peer connection is a triage lead, not proof of this exploit.
02 / REQUEST
What did it ask to do?
A decoded request can support a mechanism-specific assessment. The service described by Horizon3 uses TLS; ordinary flow logs cannot reveal the request body. Payload inspection needs appropriate TLS visibility and protocol decoding.
03 / HOST CHANGE
What changed or ran?
Correlate file content, execution evidence, and subsequent activity. A suspicious request does not by itself prove a successful write or command execution.
Horizon3 did not find application logs explicitly identifying exploitation in its research. Its file-investigation leads include /opt/charting/redishb.sh and /opt/phoenix/bin/phLicenseTool.
Compare suspicious files with a trusted baseline for the same build. Preserve the original content, hash, relevant metadata, collection time, and collection method. Explain legitimate maintenance before assigning malicious intent. Conversely, an unchanged timestamp or an empty search is not enough to close the case; record which sources were available and how far back they reached.
Give the evidence its own access boundary
A second copy is useful only if the potentially compromised appliance cannot quietly rewrite or remove it. Check the access model, not just the storage location. Separate credentials, retention controls, and a tested retrieval process matter more than calling the destination “off-host.”
Start with records you already have: firewall activity, infrastructure audit trails, administrative changes, and externally retained telemetry. Use supported collection methods and agree on operational impact with the platform owner.
Test retrieval in a controlled exercise: can another analyst obtain the retained record without signing into the SIEM? Can they establish its host, time window, source, collection method, and gaps?
Where deception fits
A bounded decoy can test whether unexpected service access reaches an independent alert and an accountable responder. A plain port listener tests connection visibility; request-level testing requires a suitable protocol implementation and safe fixtures. Isolate it from production, and define its owner, expected benign traffic, stop condition, and independently controlled evidence destination.
This is a proposed validation exercise; we have not validated FortiSIEM exploit coverage. For deployments alongside OT, map actual collection paths and privileges rather than assuming that receiving industrial logs grants control-network access.
USE THIS IN THE HANDOFF
Close two different questions
Is the exposure addressed?
- Exact version and vendor applicability decision.
- Supported update, completed change, and verification.
- Required peers and tested access restrictions.
Is the incident question answered?
- Changes and activity investigated.
- Evidence retained independently and retrievable.
- Search window, missing telemetry, and unresolved scope.
For each open item: name the owner, next action, and due time. A successful update closes a patch task. It does not explain earlier activity.
Make the handoff usable
Leave the next responder with a clear question, the records needed to investigate it, and an action. That is what we mean by earlier warning and better evidence.
Download the response carouselRead the accessible text version
Sources and scope
Checked October 22, 2026. Vulnerability details come from Fortinet and Horizon3; the response workflow is Deception Check's analysis. We did not execute the PoC or collect first-party exploit captures. Horizon3's narrative describes cron, while its current source names phxctl.service; we use “recurring execution” to cover that unresolved difference. Refresh the sources before acting.
