A flaw scored 6.0 carried a root implant onto FortiGate firewalls
CVE-2025-61624 is an authenticated path traversal rated medium by its vendor and by NVD. Unit 42 records it as the step that turned stolen management credentials into a memory-resident implant. CISA catalogued it on 16 October with a 19 October federal deadline.

A medium score and a root implant
On 14 October 2026 Palo Alto Networks Unit 42 published an account of a prolonged intrusion at a telecommunications managed service provider and of an implant it names PEERSTITCH [1]. Two days later CISA added CVE-2025-61624 to its Known Exploited Vulnerabilities catalogue with a federal remediation date of 19 October 2026 under Binding Operational Directive 22-01 [2].
CVE-2025-61624 is a path traversal in the command line interpreter of FortiOS, FortiPAM, FortiProxy and FortiSwitchManager. Fortinet’s advisory FG-IR-26-122, published 14 April 2026, rates it medium with a CVSS v3 base score of 6.0 and a temporal score of 5.4, over a local attack vector requiring high privilege [3]. The National Vulnerability Database scores the same defect 6.5 over a network vector [4]. One outlet reported it as 5.4 [5]. Siemens, whose RUGGEDCOM APE1808 carries the Fortinet firewall component, reproduces the 6.0 figure in advisory SSA-975644 [6]. Four published numbers for one flaw, three of them different.
Each score is defensible on its own terms. None of them describes the role the flaw played. In this intrusion the privilege that the score discounts had already been taken from somewhere else.
Where the privilege came from
Unit 42 records initial access through CVE-2024-47575 on FortiManager, the platform that manages the firewalls [1]. Fortinet’s advisory FG-IR-24-423, published 23 October 2024 and updated that November, rates that flaw critical at 9.8 for missing authentication on a critical function, allowing a remote unauthenticated attacker to run code or commands [7]. Mandiant, publishing the following day, tracked the activity as UNC5820, placed earliest observed exploitation at 27 June 2024, and identified more than fifty potentially compromised FortiManager devices across several industries [8]. The exfiltrated archive held configurations for managed firewalls, their serial numbers and addresses, user details and hashed passwords [8]. Mandiant stated that it lacked sufficient data to assess actor motivation or location [8].
In the case Unit 42 describes, the stolen archive covered more than 500 managed firewalls, including configurations, serial numbers, addresses, and credential and key material [1]. Security Affairs renders the figure as about 500 [2]; SC World and The Hacker News as more than 500 [5][9]. Unit 42 says it did not confirm intrusion into the downstream client networks themselves [1].
So by the time the medium-severity flaw was used, the attacker held administrative credentials recovered from the management platform and had opened remote access sessions with them [1]. An authenticated, high-privilege flaw is a different proposition when the authentication was taken two years earlier. CVSS prices the privilege requirement as a barrier. On a management appliance whose function is to hold other systems’ secrets, it is not a barrier; it is the appliance’s purpose turned around.
Why the appliance cannot tell you
The implant is memory-resident. Unit 42 describes a 64-bit ARM binary written in C that patches two long-running FortiOS processes, the logging daemon and the key-exchange daemon, and passes data between them through process memory rather than through sockets or files. It writes no persistent binary to disk, opens no listening port, and makes no outbound connection of its own [1][5]. A separate channel, abusing a certificate-management module on the attacker’s web server, sustained a reverse shell [1][9].
This matters because the device is closed. An endpoint agent cannot be installed on it, and the checks that are available, which is to say firmware integrity verification, configuration review, listening-port enumeration and outbound connection review, are precisely the checks this design is built to pass. Unit 42’s own detection guidance is therefore memory inspection and crash-log review on the appliance itself [1].
Anti-forensics compounds it. Unit 42 records that logging was stopped, crash logs cleared, configuration rolled back and decoy files planted [1]. A defender reading only the appliance’s own record would find a clean device and close the ticket.
The report is not internally consistent on every detail: its narrative and its indicator table give different sizes for the injected payload [1]. Minor, and noted because single-source technical accounts should be read with their inconsistencies visible.
The timeline is the second finding
Set the dates in order. Unit 42 places the start of the intrusion in September 2024, against a FortiManager flaw disclosed publicly in October 2024 [1][7]. The escalation flaw was published on 14 April 2026 [3]. The National Vulnerability Database record was last modified on 17 June 2026, with no exploitation note at that point [4]. The catalogue addition came on 16 October 2026 with a deadline three days later [2].
Two years separate the intrusion from the public record of the flaw that carried it, and six months separate the patch from the exploitation notice. Organisations that triage by catalogue membership were not told to prioritise this one until this week. Organisations that triage by severity would have ranked it low: the MS-ISAC advisory of 14 April 2026 listed it among more than twenty Fortinet identifiers, with five others marked as the most severe in the set [10].
In October 2024 Tenable recorded, from internet scan data, nearly 60,000 internet-facing FortiManager devices, more than 13,000 of them in the United States and 2,300 in India [11]. That is the population from which management-plane credential theft draws.
This is also not an isolated pattern. Mandiant found a separate zero-day in another vendor’s remote support appliance in late September 2026 while investigating an unrelated edge-device intrusion [12]. The management plane is where two independent investigations landed in the same month.
What to do
- Patch to the fixed releases named in FG-IR-26-122: FortiOS 7.6.5 or 7.4.10, FortiProxy 7.6.5 or 7.4.12, FortiSwitchManager 7.2.8 or 7.0.7, FortiPAM 1.7.1, and migrate off the branches the advisory marks as affected in all versions [3]. Owners of the Siemens APE1808 take the vendor route in SSA-975644 rather than patching the component directly [6].
- Treat any historical FortiManager exposure to CVE-2024-47575 as credential loss on every managed device and rotate accordingly. Fortinet’s own 2024 advisory says exactly this, and it was issued two years ago [7].
- Do not accept a clean appliance self-report as evidence of a clean appliance. Use the vendor-published memory and crash-log checks, and capture them before any reboot, firmware reinstall or configuration reset, because each of those destroys the only state the implant occupies [1].
- Stop ranking appliance flaws on base CVSS alone. A privilege prerequisite the environment has already conceded is not a mitigating factor, and a local attack vector on a device reached through a management protocol is not a local attack.
- Where the firewall estate is managed by a third party, ask for the management platform’s patch and credential-rotation history in writing. The intrusion Unit 42 describes began at a service provider and reached its customers’ device configurations [1].
The OFFSEC read
Attribution is reported, not asserted here. Unit 42 tracks the cluster as CL-STA-0954, also known as UNC5820, and records that parts of the post-exploitation activity overlap with intrusions attributed to suspected nation-state actors, without making a firm attribution [1]. Mandiant’s position in 2024 was that it could not assess motivation or location [8]. Those are the researchers’ assessments and they are the limit of what has been published.
The transferable lesson is about scoring, not about Fortinet. Vulnerability severity frameworks model an attacker who starts outside. On security appliances the realistic attacker starts with the appliance’s own credentials, because an earlier flaw in the platform above it handed them over. Until severity scoring accounts for that, a 6.0 will keep being deprioritised in environments where it is the last step rather than the first.
Sources
Every R3KONX article cites its primary material. 12 sources, in order of first citation. Links open the original publication.
- Unit 42 Uncovers PEERSTITCH, a Stealthy FortiGate Implant Palo Alto Networks Unit 42 · 2026-10-14
- U.S. CISA adds Fortinet Multiple Products, and BeyondTrust Remote Support flaws to its Known Exploited Vulnerabilities catalog Security Affairs · 2026-10-17
- Path Traversal in CLI (FG-IR-26-122, CVE-2025-61624) Fortinet PSIRT · 2026-04-14
- CVE-2025-61624 Detail NIST National Vulnerability Database · 2026-06-17
- PEERSTITCH Malware Deployed via FortiOS Flaw Bypasses Checks on Tagged VPN Sessions The Hacker News · 2026-10-16
- SSA-975644: Multiple Vulnerabilities in Fortigate NGFW on RUGGEDCOM APE1808 Devices Siemens ProductCERT · 2026-05-12
- Missing Authentication in fgfmsd (FG-IR-24-423, CVE-2024-47575) Fortinet PSIRT · 2024-11-27
- Investigating FortiManager Zero-Day Exploitation (CVE-2024-47575) Mandiant / Google Cloud · 2024-10-24
- New PEERSTITCH malware exploits Fortinet flaw for covert communication SC World · 2026-10-16
- Multiple Vulnerabilities in Fortinet Products Could Allow for Arbitrary Code Execution (2026-035) Center for Internet Security / MS-ISAC · 2026-04-14
- CVE-2024-47575: Frequently Asked Questions About FortiJump Zero-Day in FortiManager and FortiManager Cloud Tenable · 2024-10-23
- China-nexus espionage actor targets BeyondTrust support tool in zero-day campaign Cybersecurity Dive · 2026-10-16
Researched and written by the R3KONX analysis desk from the cited primary material. Methodological caveat: the intrusion timeline, implant behaviour and victim scope are reported from a single incident-response account by Palo Alto Networks Unit 42 and have not been corroborated by a second responder; the CISA catalogue addition and federal remediation date are cited through secondary reporting because the catalogue pages could not be retrieved. No indicators, file paths or exploitation detail are reproduced here. Corrections to event@r3konx.asia.
