The exploitation clock is no longer measured in weeks. For organizations running Fortinet's FortiSandbox, it is now measured in days — and in at least one alarming case, the attack surface opened before administrators could realistically respond. Three critical vulnerabilities affecting FortiSandbox infrastructure are being actively weaponized in the wild, including one whose patch was released only days before attackers began leveraging it.
Three Flaws, One Concentrated Campaign
Security researchers confirmed observing simultaneous exploitation attempts against three distinct vulnerabilities within a 24-hour period — an unusually concentrated burst of activity that signals organized, opportunistic threat actors scanning for exposed Fortinet infrastructure at scale.
Two of the vulnerabilities carry a CVSS severity score of 9.1, placing them firmly in the critical tier. The first is a path traversal flaw within FortiSandbox's JRPC API, enabling an unauthenticated attacker to bypass authentication entirely through crafted HTTP requests. The second is an OS command injection vulnerability sharing the same attack vector and impact profile — unauthenticated remote code execution via malformed HTTP traffic. Both flaws had vendor-issued patches available for approximately two months before exploitation was confirmed. That two-month window was, apparently, insufficient for a meaningful portion of the installed base to act.
The third vulnerability broadens the attack surface considerably. An OS command injection flaw affecting not only on-premises FortiSandbox deployments but also cloud and PaaS web interface variants, it too allows unauthenticated attackers to execute arbitrary system commands through specifically crafted HTTP requests. This particular flaw received its patch just one week before active exploitation was observed in the wild.
AI-Generated Exploits: Crude, Buggy — and Still Working
Perhaps the most unsettling detail to emerge from analysis of this campaign is the nature of the exploit targeting the most recently patched vulnerability. Researchers examining the exploit code believe it bears the hallmarks of AI-assisted development — and not sophisticated AI-assisted development. The code is reportedly flawed, containing bugs that suggest a rushed or automated generation process with limited quality review.
Attackers are throwing broken AI-generated exploit code at unpatched enterprise systems — and still achieving results. That outcome is less a testament to attacker sophistication and more an indictment of the patch management posture across the broader Fortinet user base.
No public proof-of-concept exploit for this vulnerability had been released at the time of confirmed exploitation. This means threat actors were not simply repurposing community research — they were independently developing functional-enough attack tools, likely with AI assistance, fast enough to strike before defenders could patch. The fact that a buggy, AI-generated exploit can still find purchase against production systems is a damning commentary on the state of enterprise patch hygiene.
A Pattern Fortinet Cannot Ignore
This incident does not exist in isolation. Fortinet's product portfolio has faced a sustained pattern of rapid post-disclosure exploitation. Earlier this year, a critical vulnerability in FortiClient EMS — also carrying a CVSS score of 9.1 — was being actively exploited before the vendor had even issued a remediation. Out-of-band emergency patches followed, but the damage window had already opened.
Taken together, these incidents paint a picture of Fortinet infrastructure as a high-value, high-return target class for threat actors. The combination of widespread enterprise deployment, privileged network positioning, and — critically — slow organizational patch adoption makes Fortinet appliances a reliable hunting ground. Attackers have clearly internalized this calculus.
What You Should Do
The Defender's Shrinking Window
The fundamental calculus of vulnerability management has shifted. For years, security teams operated under a mental model in which critical patches required deployment within 30 days, with 90 days representing an outer boundary for less severe issues. That model is obsolete for network security appliances and widely deployed enterprise platforms. The window between patch release and active exploitation is now frequently measured in single-digit days — and in this case, for one vulnerability, in hours.
Organizations that have not yet operationalized rapid patch deployment pipelines for their security infrastructure are not running a calculated risk. They are simply waiting to become a statistic. The FortiSandbox campaign is a clear signal: the adversary timeline has accelerated, and enterprise security operations must accelerate with it.
