CVE-2025-50472
CriticalSummary
Biblioteka modelscope/ms-swift w wersjach do 2.6.1 jest podatna na wykonanie dowolnego kodu poprzez deserializację niezaufanych danych w funkcji `load_model_meta()` klasy `ModelFileSystemCache()`. Atakujący mogą wykonać dowolny kod, tworząc złośliwy, serializowany ładunek `.mdl`, co umożliwia zdalne wykonanie kodu (RCE).
Risk Assessment
Podatność ta stwarza poważne ryzyko dla organizacji, ponieważ atakujący mogą zdalnie wykonywać kod na zainfekowanych maszynach, co może prowadzić do utraty danych lub przejęcia kontroli nad systemem. Dodatkowo, złośliwy ładunek jest ukryty, co utrudnia jego wykrycie przez ofiary.
Recommendation
Zaleca się aktualizację biblioteki modelscope/ms-swift do najnowszej wersji, aby usunąć tę podatność. Należy również przeprowadzić audyt kodu i monitorować procesy związane z ładowaniem modeli, aby zminimalizować ryzyko wykonania złośliwego kodu.
Related vulnerabilities
- CVE-2026-109157Critical
The TRENDnet TEW-432BRP router version 3.10B20 contains a stack-based buffer overflow in the formTcpipSetup function of /goform/formTcpipSetup. Manipulating the Name argument allows remote attack, and the exploit has been publicly disclosed.
- CVE-2026-63354Critical
Affected devices do not properly validate and sanitize visual marker data contained in imported CSV trace files before rendering it on the Trace page. An attacker can socially engineer an authorized user with the 'Read diagnostics' right into importing a crafted trace file.
- CVE-2026-63353Critical
Affected devices do not properly sanitize comment fields contained in imported trace files before rendering them on the Trace page. An attacker can socially engineer an authorized user with the 'Read diagnostics' right into importing a crafted trace file.
- CVE-2026-63350Critical
Affected devices do not properly restrict executable content contained in imported trace files. An attacker can socially engineer an authorized user with the 'Read diagnostics' right into importing a crafted trace file.
- CVE-2026-44415Critical
Affected systems do not properly restrict access to internal APIs and paths for authenticated users. An authenticated remote attacker could access sensitive internal information.
- CVE-2026-109110Critical
The Wavlink WL-WN592AX6 router version 250422 contains a vulnerability in the system function of the wmapd daemon component. Manipulation of an argument leads to OS command injection, and the attack can be launched remotely.
- CVE-2026-109109Critical
The Wavlink WL-WN530BE1-A router version 251117 contains a stack-based buffer overflow in the strcpy function of /cgi-bin/ExportAllSettings.cgi in the Settings Export CGI component. Manipulating the HTTP_COOKIE/X-Cmd argument allows remote attack.
- CVE-2026-109108Critical
The Wavlink WL-WN530BE1-A router version 251117 contains a vulnerability in the popen function of /cgi-bin/uploadVpn.cgi. Manipulating the Type argument leads to OS command injection, and the attack can be performed remotely.
- CVE-2026-108971Critical
A flaw has been found in D-Link DCS-930L B1.08B4 affecting the mpfd_decode function in bootloader_region.bin of the Emergency HTTP Recovery Server component. Manipulating the firmware argument leads to improper verification of the cryptographic signature. The attack can be performed remotely and the exploit has been published.
- CVE-2026-109758Critical
Perforce Helix Authentication Service (HAS) versions 2022.1-2025.2 retain a default SCIM bearer token that is not automatically replaced with a random value during installation. An attacker knowing this token can gain unauthorized access to the SCIM API and perform user and group operations. This issue can be chained with CVE-2026-61834, potentially leading to remote code execution.
Original NVD description (English source)
The modelscope/ms-swift library thru 2.6.1 is vulnerable to arbitrary code execution through deserialization of untrusted data within the `load_model_meta()` function of the `ModelFileSystemCache()` class. Attackers can execute arbitrary code and commands by crafting a malicious serialized `.mdl` payload, exploiting the use of `pickle.load()` on data from potentially untrusted sources. This vulnerability allows for remote code execution (RCE) by deceiving victims into loading a seemingly harmless checkpoint during a normal training process, thereby enabling attackers to execute arbitrary code on the targeted machine. Note that the payload file is a hidden file, making it difficult for the victim to detect tampering. More importantly, during the model training process, after the `.mdl` file is loaded and executes arbitrary code, the normal training process remains unaffected'meaning the user remains unaware of the arbitrary code execution.
Vulnerability data from NVD (NIST) · CISA KEV · EPSS