An unauthenticated SQL injection vulnerability in FortiClient EMS 7.4.4 is currently under active exploitation. Over just a few days, 51 distinct IP addresses have attempted to exploit this vulnerability, as identified by the CrowdSec network. The availability of public proof-of-concept code and its registration in CISA’s Known Exploited Vulnerabilities (KEV) catalog make remediation urgent.
Designated as CVE-2026-21643 with a CVSS score of 9.1, this vulnerability impacts the centralized administration console for Fortinet endpoints, which serves as the command hub for IT teams managing endpoint protection, VPN profiles, and compliance policies across their networks. Fortinet issued an official advisory earlier this year. For several weeks, the vulnerability remained theoretically risky—known, documented, yet with no significant exploitation detected. This changed when CISA included CVE-2026-21643 in its KEV catalog. Just two days later, CrowdSec released a detection rule, and within five days following this publication, the first exploitation attempts were detected by their collaborative telemetry. This sequence underscores a well-established pattern: KEV listing acts as a signal to a segment of the offensive community, regardless of how long it has been since the initial disclosure.
FortiClient EMS serves as Fortinet’s centralized endpoint management platform. It consolidates the configuration of endpoint agents, distribution of security policies, VPN profiles, compliance settings, and inventory management. This functional positioning makes it a compelling target, distinct from standard exposed services. Compromising the EMS provides access to the control operations of endpoints and potentially to sensitive identity and remote access configuration data.
Unauthenticated Injection in an HTTP Header
The technical mechanism behind CVE-2026-21643 is remarkably straightforward. The flaw exists in how the HTTP Site header is processed at the endpoint /api/v1/init_consts of the FortiClient EMS administration interface. This value is forwarded without adequate sanitization to the backend database engine, allowing an attacker to inject arbitrary SQL commands via a simple HTTP request without presenting any authentication token.
Fortinet’s advisory (FG-IR-25-1142) classifies the vulnerability as an unauthenticated SQL injection. Research published by the expert 0xBlackash, accompanied by a public proof-of-concept (PoC) code, and a template by ProjectDiscovery converge on the same attack vector: a single request, a single manipulated header, leading to direct access to the persistence layer. Exploitation can yield various effects based on the target database’s configuration, including data extraction, record manipulation, triggering errors that reveal information useful for reconnaissance, or even chaining to remote code execution if the database engine capabilities and server configuration permit. The discovery is credited to Gwendal Guégniaud from Fortinet’s Product Security team.
The simultaneous availability of a PoC and an operational Nuclei template significantly lowers the skill level required for exploitation. Any attacker capable of launching an automated scanning campaign can identify vulnerable instances exposed on the internet and attempt exploitation without needing specialized technical knowledge.
51 Attacking IPs in One Week: A Phase of Selective Targeting
During the week following the detection of the initial attempts, CrowdSec recorded 51 distinct IP addresses engaged in exploitation attempts of CVE-2026-21643. CrowdSec characterizes the current phase as measured targeting, distinguishing it from broad campaigns that mobilize hundreds or thousands of IPs in just a few days. This relative intensity does not imply a low risk; rather, it indicates that the vulnerability has not yet been integrated into mass exploitation tools, but that organized actors are actively probing the accessible attack surface.
The window between the publication of the detection rule and the first signal of exploitation was just five days. For security teams, this timeframe represents the gap between awareness of a risk and facing active attempts. In the operational reality of businesses—considering validation cycles, regression testing, and resource availability—five days is a challenging constraint to navigate for tightly scheduled administration systems.
Affected Versions and Scope of Remediation
According to Fortinet’s advisory, only version 7.4.4 of FortiClient EMS is affected. Versions 7.2.x and 8.0.x are not within the scope of the vulnerability. The official corrective measure is to update to version 7.4.5 or higher.
For environments unable to patch immediately, CrowdSec recommends removing the EMS administration interface from direct internet exposure. Access should be restricted via a VPN, an identity-aware proxy, or a strict allowlist. Implementing a web application firewall (WAF) with request inspection targeting /api/v1/init_consts and monitoring for abnormal Site header values can provide an additional detection layer.
The gap between the initial disclosure and the first observed exploitation attempts is 73 days. This figure can be misleading; it does not reflect a slow response time from attackers, but rather the delayed trigger effect of KEV registration. For security teams, the operational lesson is clear: patch prioritization must account for changes in vulnerability status in reference catalogs and the availability of public exploitation material, just as much as it considers the age of the initial disclosure.
The combination of a CVSS score of 9.1, an unauthenticated vector, public PoC availability, and confirmed exploitation across 51 distinct IPs within a week surpasses the threshold that justifies waiting for a scheduled patch cycle. Organizations using FortiClient EMS 7.4.4 must treat CVE-2026-21643 as an immediate remediation priority.



