Could Your SASE Client Grant SYSTEM Access?

Article Highlights
Off On

A recently disclosed vulnerability highlights a critical risk within enterprise security stacks, demonstrating how a trusted software agent designed to protect a system can inadvertently become the very key an attacker needs for a complete takeover. A severe privilege escalation flaw, identified as CVE-2025-9142, was discovered within the Check Point Harmony SASE Windows client, affecting all versions prior to 12.2. This vulnerability allowed a local attacker with standard user permissions to write or delete files in arbitrary system locations, a capability that serves as a direct pathway to executing malicious code with the highest possible privileges. The discovery serves as a stark reminder that even sophisticated security solutions are not immune to fundamental programming errors, and a single misstep in validating user-supplied data can unravel an entire system’s defenses, turning a protective tool into a powerful weapon for an intruder. The flaw’s mechanics reveal a classic yet potent attack chain, leveraging a trusted, high-privilege service against itself.

The Anatomy of a Token-Based Exploit

At the heart of this critical vulnerability lies the Perimeter81.Service.exe component, a core process of the Harmony SASE client that operates with SYSTEM-level permissions, granting it unrestricted access to the local machine. The fundamental flaw was rooted in an insufficient validation of JSON Web Tokens (JWT) during the client’s authentication sequence. When a user initiates a login, the client utilizes a custom URI handler, perimeter81://, to pass a JWT to this privileged service. Under normal circumstances, the service should rigorously verify the token’s digital signature to ensure its authenticity and integrity before processing its contents. However, investigations revealed that this crucial verification step was either missing or improperly implemented. This oversight created a significant security loophole, allowing an unauthenticated local attacker to craft a malicious URL containing a tampered JWT and pass it directly to the service. Because the signature was not checked, the service blindly trusted the manipulated token, processing its malicious payload as if it were a legitimate request from an authenticated user.

The exploitation of this token validation flaw unfolds in a carefully orchestrated, two-phase attack. In the initial phase, an attacker leverages their ability to forge a JWT to achieve an arbitrary directory creation primitive. By injecting directory traversal sequences—such as ../../../—into the token’s tenant ID field, the attacker could trick the service into creating a folder structure outside of its intended working directory. To bypass an internal client-side check, this attack required the use of a domain present on a whitelist; however, it was noted that one such domain, p81-falcon.com, was available for public registration, removing a significant barrier. Following this, the second phase involves a sophisticated symbolic link injection technique. Using native Windows tools like Object Manager and RPC Control, the attacker can create specific symlinks that redirect file operations. They then trigger the GenerateAndLoadCertificates() function within the privileged service. This function, which runs as SYSTEM, is then fooled by the symlinks into writing a file, such as a client certificate, to a protected system directory like C:WindowsSystem32 instead of its intended location.

From Arbitrary File Write to Full System Control

The ability to write a file to any location on the system is a powerful primitive that attackers can readily convert into full system compromise, and in this case, the most direct path was through DLL hijacking. Analysis of the Perimeter81.Service.exe process revealed that upon startup, it attempts to load several Dynamic-Link Libraries (DLLs) from its working directory. This behavior is common but becomes a severe security risk when an attacker can control the contents of that directory. By combining the arbitrary file write capability achieved through the JWT and symbolic link attack, an adversary could place a malicious DLL with a specific, expected name into a location where the service would look for it. When the service is next restarted, either by the system or triggered by the attacker, it would load and execute the code within the malicious DLL. Since the service itself runs with SYSTEM privileges, the attacker’s code would inherit those same permissions, effectively completing the privilege escalation from a standard user to the highest level of access on the machine.

Check Point was first notified of this significant security flaw on March 16, 2025, and after a period of development and testing, a patched version of the Harmony SASE client was released. The fix, included in version 12.2 and all subsequent releases, was made available to the public on November 18, 2025. The patch addresses the root cause of the vulnerability by implementing proper and stringent validation of the JWT, ensuring that the service now correctly verifies the digital signature of any incoming token before processing its payload. This prevents an attacker from being able to pass a tampered token to the service, thereby neutralizing the entire attack chain. For organizations deploying this SASE solution, the primary and most critical mitigation step was to ensure an immediate upgrade of all Windows clients to the patched version. As an additional layer of defense, security teams were also advised to enforce policies that restrict local administrative access where possible and to implement application whitelisting solutions, which could further harden systems against DLL hijacking attempts.

Explore more

Is the Mistic Backdoor Hiding in Your Security Tools?

Introduction The emergence of the Mistic backdoor represents a sophisticated advancement in the arsenal of modern cybercriminals, specifically those operating within the niche of Initial Access Brokering (IAB). This malicious software, also identified by some security researchers as MLTBackdoor, has been actively infiltrating corporate environments throughout the first half of 2026. Its primary strength lies in its ability to camouflage

Is the Redmi 17C the New King of Budget Smartphones?

Dominic Jainy is a seasoned IT professional with a deep understanding of how hardware evolution impacts the budget mobile market. Today, he breaks down Xiaomi’s latest strategic move with the Redmi 17C, a device that surprisingly leaps over a generation to deliver high-refresh-rate displays and massive battery life to the entry-level segment. We explore the balance between essential utility features,

How Can PowerTool Speed Up Business Central Data Migrations?

Modern enterprises frequently encounter significant friction during ERP transitions because traditional data migration methods often fail to accommodate the sheer volume and complexity of contemporary datasets. In 2026, the demand for agility within Microsoft Dynamics 365 Business Central has reached a point where standard configuration packages, while functional for small tasks, often act as a bottleneck for larger implementations. The

How to Move Beyond the Portal to a True Developer Platform?

Dominic Jainy stands at the forefront of the modern cloud-native movement, possessing a deep technical mastery of artificial intelligence, machine learning, and blockchain architectures. With years of experience navigating the complexities of large-scale IT infrastructures, he has become a leading voice in the evolution of platform engineering. His perspective is shaped by the practical realities of moving beyond simple automation

Will AI Token Costs Soon Surpass Developer Salaries?

Recent financial projections indicate that the cost of maintaining high-frequency artificial intelligence interactions is rapidly approaching the median annual compensation of experienced software engineers in the global market. As the software development industry undergoes a radical transformation, the traditional overhead associated with human labor is being challenged by the sheer volume of data processed through large language models. This shift