NEWSkillSec — elevating AI Skills security from malware detection to capability auditingSkillSecLearn more →
← Back to Intel Center
SECURITY INTEL · DAILY
RSS

2026-10-12 Daily Security Intelligence

CSSA 2 · CVE 2 · Poison 0

10.0 Critical · CVE-2026-108576 Command injection in TOZED X300 IPPingDiagnosti…

CSSA Exclusive Early Warning 2

CSSA Exclusive Early Warning 9.9 Critical

Authorization logic flaw in sglang lets ordinary credentials perform admin operations and affect model integrity

When only an ordinary API key is configured and no dedicated admin key is set, the authentication middleware incorrectly treats the ordinary key as a valid admin credential when evaluating ADMIN_OPTIONAL endpoint permissions. An attacker can exploit this logic flaw by sending malicious requests that carry the ordinary key over the network, then access sensitive endpoints such as /update_weights_from_disk to illegally modify model weights, load arbitrary LoRA adapters, or start profiling, harming model data integrity and service availability.

Component
sglang is a high-performance large language model inference serving engine built with a mix of Python and C++. It aims to increase throughput through optimized memory management and parallel processing. Core components receive HTTP requests, schedule GPU compute resources, and manage model weight loading for AI application deployments that need low latency and high concurrency.
Type
Incorrect Authorization (CWE-863)
Repo
github.com/sgl-project/sglang · Stars 37023
Remediation
  • Enforce strict role separation between user and admin credentials, remove implicit privilege elevation for single-key deployments at the code level, and require independent high-strength authentication for all admin operations.
CSSA Exclusive Early Warning 9.6 Critical

Input validation flaw in ProxySQL leaks session state and breaks multi-tenant data isolation

An attacker exploits input validation failure caused by a matching regular expression that lacks boundary anchors. By constructing non-standard GUCs whose names are prefixed with controlled variable names (for example timezone_abbreviations) and issuing SET commands, the request enters a parse-failure path without locking the host group or disabling multiplexing. Session variables then silently remain in the connection pool and leak to other clients, causing cross-session data confusion and integrity damage.

Component
ProxySQL is a high-performance MySQL and PostgreSQL middleware. Its core architecture uses connection pooling and query routing to optimize backend database load through session reuse, enabling resource scheduling, load balancing, and failover in high-concurrency enterprise distributed database infrastructure.
Type
Improper Input Validation (CWE-20)
Repo
github.com/sysown/proxysql · Stars 6936
Remediation
  • Fix the regular expression to add strict suffix or word-boundary checks so uncontrolled variables trigger host-group locking. When handling parse failures, distinguish unknown variables from format errors, and force a safe forwarding mode for unknown variables to prevent accidental session-state sharing.

CVE Intelligence 2

CVE-2026-108576 CVSS 10.0 Critical 2026-10-11

Command injection in TOZED X300 IPPingDiagnostics leads to remote code execution

In TOZED X300 version 6.01.3 and earlier, the process_ping function in the IPPingDiagnostics Handler component lacks effective filtering of the Host parameter, resulting in an OS command injection (CWE-77) flaw. An attacker can inject malicious system commands by manipulating the Host parameter and execute arbitrary code on the target device. The vulnerability allows remote attackers to attack without authentication and severely threatens confidentiality, integrity, and availability. All TOZED X300 users who have not updated to a secure version are affected. An attacker only needs to send a crafted request over the network; no user interaction is required.

Component
TOZED X300 is a network device whose IPPingDiagnostics Handler component processes Ping requests related to network diagnostics.
Risks
  • Complete system control: An attacker can execute arbitrary code on the victim system and, depending on privileges, install programs, view/change/delete data, or create new fully privileged accounts
  • No user interaction: The vulnerability can be triggered by a remote network request without any additional action by the target user
  • High-impact severity: A CVSS score of 10.0 indicates severe impact on confidentiality, integrity, and availability
Source
NVD · CVE-2026-108576
Remediation
  • Deploy a firewall or intrusion detection system at the network boundary to block anomalous traffic targeting the IPPingDiagnostics port
  • Restrict access to diagnostic features so that only trusted administrator IP addresses are allowed
CVE-2026-108804 CVSS 10.0 Critical 2026-10-12

Remote command injection in TRENDnet router pigwidgeon.cgi leads to remote code execution

In TRENDnet TEW-751DR, TEW-752DRU, and TEW733GR firmware versions 1.03B01/1.03B03, the Web Management Backend component lacks input validation and contains a command injection (CWE-77) flaw. Manipulating the service parameter in pigwidgeon.cgi can trigger OS command injection and threaten device integrity and availability. Attackers can craft malicious requests against the Web management backend to achieve remote code execution or full device control. All users of affected TRENDnet router firmware are impacted. The attack can be launched remotely without user interaction, and exploit code is publicly available.

Component
TRENDnet routers are network hardware devices that provide a Web management backend for configuring and managing network settings.
Risks
  • Complete system control: An attacker can execute arbitrary system commands on the router and, depending on privileges, modify configuration, view/change/delete data, or install malware
  • No user interaction: Through a remote network request (T1190), an attacker can trigger the vulnerability and control the device without user interaction
  • Remote exploitation: Public exploit code may be widely used by automated attack tools, leading to large-scale device compromise
Source
NVD · CVE-2026-108804
Remediation
  • Immediately stop using end-of-life (EOL) TRENDnet TEW-751DR, TEW-752DRU, and TEW733GR routers
  • Replace them with current models that receive ongoing vendor security support
  • Deploy firewall rules at the network boundary to block unauthorized access to the router Web management interface
← Back to Intel CenterRSSMonitor these risks automatically with CleanSource SCA →