| CVE |
Vendors |
Products |
Updated |
CVSS v3.1 |
| vLLM is an inference and serving engine for large language models (LLMs). Starting in version 0.10.1 and prior to version 0.18.0, two model implementation files hardcode `trust_remote_code=True` when loading sub-components, bypassing the user's explicit `--trust-remote-code=False` security opt-out. This enables remote code execution via malicious model repositories even when the user has explicitly disabled remote code trust. Version 0.18.0 patches the issue. |
| A flaw was found in Undertow. A remote attacker can exploit this vulnerability by sending `\r\r\r` as a header block terminator. This can be used for request smuggling with certain proxy servers, such as older versions of Apache Traffic Server and Google Cloud Classic Application Load Balancer, potentially leading to unauthorized access or manipulation of web requests. |
| A flaw was found in Undertow. This vulnerability allows a remote attacker to construct specially crafted requests where header names are parsed differently by Undertow compared to upstream proxies. This discrepancy in header interpretation can be exploited to launch request smuggling attacks, potentially bypassing security controls and accessing unauthorized resources. |
| A flaw was found in Undertow. When Undertow receives an HTTP request where the first header line starts with one or more spaces, it incorrectly processes the request by stripping these leading spaces. This behavior, which violates HTTP standards, can be exploited by a remote attacker to perform request smuggling. Request smuggling allows an attacker to bypass security mechanisms, access restricted information, or manipulate web caches, potentially leading to unauthorized actions or data exposure. |
| Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to version 0.8.6, any authenticated user can overwrite any file's content by ID through the `POST /api/v1/retrieval/process/files/batch` endpoint. The endpoint performs no ownership check, so a regular user with read access to a shared knowledge base can obtain file UUIDs via `GET /api/v1/knowledge/{id}/files` and then overwrite those files, escalating from read to write. The overwritten content is served to the LLM via RAG, meaning the attacker controls what the model tells other users. Version 0.8.6 patches the issue. |
| Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to version 0.8.6, any authenticated user can read other users' private memories via `/api/v1/retrieval/query/collection`. Version 0.8.6 patches the issue. |
| A path traversal vulnerability exists in the awesome-llm-apps project in commit e46690f99c3f08be80a9877fab52acacf7ab8251 (2026-01-19) in the Beifong AI News and Podcast Agent backend in FastAPI backend, stream-audio endpoint, in file routers/podcast_router.py, in function stream_audio. The stream-audio endpoint accepts a user-controlled path parameter that is concatenated into a filesystem path without proper validation or restriction. An unauthenticated remote attacker can exploit this vulnerability to read arbitrary files from the server filesystem, potentially disclosing sensitive information such as configuration files and credentials. |
| The command auto-approval module in CodeRider-Kilo contains an OS Command Injection vulnerability, rendering its whitelist security mechanism ineffective. The vulnerability stems from the incorrect use of an incompatible command parser (the Unix-based shell-quote library) to analyze commands on the Windows platform, coupled with a failure to correctly handle Windows CMD-specific escape sequences (^). Attackers can exploit this discrepancy between the parsing logic and the execution environment by constructing payloads such as git log ^" & malicious_command ^". The CodeRider-Kilo parser is deceived by the escape characters, misinterpreting the malicious command connector (&) as being within a protected string argument and thus auto-approving the command. However, the underlying Windows CMD interpreter ignores the escaped quotes, parsing and executing the subsequent malicious command directly. This allows attackers to achieve arbitrary Remote Code Execution (RCE) after bypassing what appears to be a legitimate Git whitelist check. |
| The command auto-approval module in Axon Code contains an OS Command Injection vulnerability, rendering its whitelist security mechanism ineffective. The vulnerability stems from the incorrect use of an incompatible command parser (the Unix-based shell-quote library) to analyze commands on the Windows platform, coupled with a failure to correctly handle Windows CMD-specific escape sequences (^). Attackers can exploit this discrepancy between the parsing logic and the execution environment by constructing payloads such as git log ^" & malicious_command ^". The Axon Code parser is deceived by the escape characters, misinterpreting the malicious command connector (&) as being within a protected string argument and thus auto-approving the command. However, the underlying Windows CMD interpreter ignores the escaped quotes, parsing and executing the subsequent malicious command directly. This allows attackers to achieve arbitrary Remote Code Execution (RCE) after bypassing what appears to be a legitimate Git whitelist check. |
| In its design for automatic terminal command execution, AI Code offers two options: Execute safe commands and execute all commands. The description for the former states that commands determined by the model to be safe will be automatically executed, whereas if the model judges a command to be potentially destructive, it still requires user approval. However, this design is highly susceptible to prompt injection attacks. An attacker can employ a generic template to wrap any malicious command and mislead the model into misclassifying it as a 'safe' command, thereby bypassing the user approval requirement and resulting in arbitrary command execution. |
| A Business Logic vulnerability exists in SourceCodester Pharmacy Product Management System 1.0 in the add-sales.php file. The application fails to verify if the requested sales quantity (txtqty) exceeds the available stock level. An attacker can manipulate the request to purchase a quantity that is significantly higher than the actual available stock. |
| A Business Logic vulnerability exists in SourceCodester Pharmacy Product Management System 1.0 in the add-stock.php file. The application fails to validate the "txtprice" and "txttotalcost" parameters during stock entry, allowing negative financial values to be submitted. This leads to corruption of financial records, allowing attackers to manipulate inventory asset values and procurement costs. |
| A blog.admin v.8.0 and before system's getinfobytoken API interface contains an improper access control which leads to sensitive data exposure. Unauthorized parties can obtain sensitive administrator account information via a valid token, threatening system security. |
| LibreChat is a ChatGPT clone with additional features. In versions 0.8.2-rc2 through 0.8.2-rc3, the SSE streaming endpoint `/api/agents/chat/stream/:streamId` does not verify that the requesting user owns the stream. Any authenticated user who obtains or guesses a valid stream ID can subscribe and read another user's real-time chat content, including messages, AI responses, and tool invocations. Version 0.8.2 patches the issue. |
| Flannel is a network fabric for containers, designed for Kubernetes. The Flannel project includes an experimental Extension backend that allows users to easily prototype new backend types. In versions of Flannel prior to 0.28.2, this Extension backend is vulnerable to a command injection that allows an attacker who can set Kubernetes Node annotations to achieve root-level arbitrary command execution on every flannel node in the cluster. The Extension backend's SubnetAddCommand and SubnetRemoveCommand receive attacker-controlled data via stdin (from the `flannel.alpha.coreos.com/backend-data` Node annotation). The content of this annotation is unmarshalled and piped directly to a shell command without checks. Kubernetes clusters using Flannel with the Extension backend are affected by this vulnerability. Other backends such as vxlan and wireguard are unaffected. The vulnerability is fixed in version v0.28.2. As a workaround, use Flannel with another backend such as vxlan or wireguard. |
| Code injection vulnerability exists in BUFFALO Wi-Fi router products. If this vulnerability is exploited, an arbitrary code may be executed on the products. |
| Traefik is an HTTP reverse proxy and load balancer. Prior to versions 3.6.11 and 3.7.0-ea.2, Traefik's Knative provider builds router rules by interpolating user-controlled values into backtick-delimited rule expressions without escaping. In live cluster validation, Knative `rules[].hosts[]` was exploitable for host restriction bypass (for example `tenant.example.com`) || Host(`attacker.com`), producing a router that serves attacker-controlled hosts. Knative `headers[].exact` also allows rule-syntax injection and proves unsafe rule construction. In multi-tenant clusters, this can route unauthorized traffic to victim services and lead to cross-tenant traffic exposure. Versions 3.6.11 and 3.7.0-ea.2 patch the issue. |
| ByteDance Deer-Flow versions prior to commit 5dbb362 contain a stored cross-site scripting vulnerability in the artifacts API that allows attackers to execute arbitrary scripts by uploading malicious HTML or script content as artifacts. Attackers can store malicious content that executes in the browser context when users view artifacts, leading to session compromise, credential theft, and arbitrary script execution. |
| Wazuh Manager authd service in wazuh-manager packages through version 4.7.3 contains an improper restriction of client-initiated SSL/TLS renegotiation vulnerability that allows remote attackers to cause a denial of service by sending excessive renegotiation requests. Attackers can exploit the lack of renegotiation limits to consume CPU resources and render the authd service unavailable. |
| calibre is a cross-platform e-book manager for viewing, converting, editing, and cataloging e-books. Prior to version 9.6.0, a Server-Side Request Forgery vulnerability in the background-image endpoint of calibre e-book reader's web view allows an attacker to perform blind GET requests to arbitrary URLs and exfiltrate information out from the ebook sandbox. Version 9.6.0 patches the issue. |
