CVE-2026-54499
HighCVSS 7.5Exploitation Probability (EPSS)
Low risk22th percentile - higher than 22% of all known CVEs
Summary
A vulnerability in the Stanza NLP library allows arbitrary code execution via a malicious .pt model file. The model loading mechanism first attempts safe torch.load with weights_only=True, but falls back to unsafe weights_only=False on an attacker-controlled pickle.UnpicklingError, enabling arbitrary code execution when the model is loaded.
Risk Assessment
An attacker can supply a crafted model file that, when loaded by a Stanza NLP pipeline, executes arbitrary code on the victim's system, leading to full compromise of the application or server.
Recommendation
Immediately update Stanza to version 1.12.2 or later. Avoid loading models from untrusted sources.
Related vulnerabilities
- CVE-2026-84834Critical
Unauthenticated PHP Object Injection vulnerability in JobSearch versions up to 3.2.0. An attacker can inject malicious PHP objects, potentially leading to arbitrary code execution.
- CVE-2026-84814Critical
Subscriber privilege escalation vulnerability in Bricksforge versions up to 3.1.8.8, allowing a subscriber to escalate privileges.
- CVE-2026-84813Critical
Unauthenticated SQL Injection vulnerability in GeoDirectory versions up to 2.8.174. An attacker can inject SQL queries, potentially leading to unauthorized database access.
- CVE-2026-84768Critical
Unauthenticated SQL Injection vulnerability in VikAppointments Services Booking Calendar versions up to 1.2.20. An attacker can inject SQL queries, potentially leading to unauthorized database access.
- CVE-2026-84753Critical
Unauthenticated PHP Object Injection vulnerability in Mail Mint versions up to 1.31.0. An attacker can inject malicious PHP objects, potentially leading to arbitrary code execution.
- CVE-2026-84238Critical
Unauthenticated Broken Access Control vulnerability in YITH Request a Quote for WooCommerce Premium versions before 4.46.0. An attacker can access functions they should not have permission to.
- CVE-2026-85216Critical
MISP contains an authentication bypass vulnerability in its LDAP and LinOTP authentication components due to insufficient validation of user-supplied credentials. An attacker knowing a valid directory user's email address could submit an empty password, which may be accepted by an LDAP server allowing unauthenticated binds. Additionally, in the LinOTP component, empty passwords could be checked against local MISP password hashes, and LDAP-provisioned accounts could be created with empty local passwords.
- CVE-2026-85183Critical
Taipy configures its socket.io server with wildcard CORS origin and credential flag enabled, allowing any web page to establish credentialed WebSocket connections to victim applications. Attackers can open socket.io sessions from arbitrary domains and invoke state variable modifications and action callbacks without CSRF protection.
- CVE-2026-85181Critical
CAT uses Java String.hashCode as the sole integrity check for session cookies without server-side keying, allowing attackers to forge valid checksums offline. Attackers can set the x-forwarded-for header to bypass IP binding validation and create admin sessions with full configuration access.
- CVE-2026-85109Critical
A vulnerability was determined in Tenda HG10 300001138. This issue affects the function formLogin of the file /boaform/formLogin of the component Boa Web Server. Executing a manipulation of the argument Username can lead to buffer overflow. The attack may be launched remotely. The exploit has been publicly disclosed and may be utilized.
Original NVD description (English source)
Stanza is a Stanford NLP Python library for tokenization, sentence segmentation, NER, and parsing of many human languages. Prior to 1.12.2, Stanza model loaders such as stanza.models.common.pretrain.Pretrain.load() attempt torch.load(..., weights_only=True) but fall back to torch.load(..., weights_only=False) on attacker-controllable pickle.UnpicklingError, allowing a malicious .pt pretrain or model file to execute arbitrary pickle code when a Stanza NLP pipeline loads it. This issue is fixed in version 1.12.2.

