vLLM is an inference and serving engine for large language models (LLMs). From version 0.5.5 to before 0.11.1, the /v1/chat/completions and /tokenize endpoints allow a chat_template_kwargs request parameter that is used in the code before it is properly validated against the chat template. With the right chat_template_kwargs parameters, it is possible to block processing of the API server for long periods of time, delaying all other requests. This issue has been patched in version 0.11.1.

In versions before 20.0.2, it was found that --disallow-code-generation-from-strings is not sufficient for isolating untrusted JavaScript in happy-dom. The untrusted script and the rest of the application still run in the same Isolate/process, so attackers can deploy prototype pollution payloads to hijack important references like "process" in the example below, or to hijack control flow via flipping checks of undefined property. This vulnerability is due to an incomplete fix for CVE-2025-61927. The vulnerability is fixed in 20.0.2.

Envoy is a cloud-native, open source edge and service proxy. Prior to 1.36.1, 1.35.5, 1.34.9, and 1.33.10, large requests and responses can potentially trigger TCP connection pool crashes due to flow control management in Envoy. It will happen when the connection is closing but upstream data is still coming, resulting in a buffer watermark callback nullptr reference. The vulnerability impacts TCP proxy and HTTP 1 & 2 mixed use cases based on ALPN. This vulnerability is fixed in 1.36.1, 1.35.5, 1.34.9, and 1.33.10.

c-ares is an asynchronous resolver library. Versions 1.32.3 through 1.34.5 terminate a query after maximum attempts when using read_answer() and process_answer(), which can cause a Denial of Service. This issue is fixed in version 1.34.6.

go-witness and witness are Go modules for generating attestations. In go-witness versions 0.8.6 and earlier and witness versions 0.9.2 and earlier the AWS attestor improperly verifies AWS EC2 instance identity documents. Verification can incorrectly succeed when a signature is not present or is empty, and when RSA signature verification fails. The attestor also embeds a single legacy global AWS public certificate and does not account for newer region specific certificates issued in 2024, making detection of forged documents difficult without additional trusted region data. An attacker able to supply or intercept instance identity document data (such as through Instance Metadata Service impersonation) can cause a forged identity document to be accepted, leading to incorrect trust decisions based on the attestation. This is fixed in go-witness 0.9.1 and witness 0.10.1. As a workaround, manually verify the included identity document, signature, and public key with standard tools (for e...

vLLM is an inference and serving engine for large language models (LLMs). From version 0.5.5 to before 0.11.1, users can crash the vLLM engine serving multimodal models by passing multimodal embedding inputs with correct ndim but incorrect shape (e.g. hidden dimension is wrong), regardless of whether the model is intended to support such inputs (as defined in the Supported Models page). This issue has been patched in version 0.11.1.

In the eap-mschapv2 plugin (client-side) in strongSwan before 6.0.3, a malicious EAP-MSCHAPv2 server can send a crafted message of size 6 through 8, and cause an integer underflow that potentially results in a heap-based buffer overflow.

A flaw was identified in the X.Org X server’s X Keyboard (Xkb) extension where improper bounds checking in the XkbSetCompatMap() function can cause an unsigned short overflow. If an attacker sends specially crafted input data, the value calculation may overflow, leading to memory corruption or a crash.

A flaw was discovered in the X.Org X server’s X Keyboard (Xkb) extension when handling client resource cleanup. The software frees certain data structures without properly detaching related resources, leading to a use-after-free condition. This can cause memory corruption or a crash when affected clients disconnect.

A flaw was found in the X.Org X server and Xwayland when processing X11 Present extension notifications. Improper error handling during notification creation can leave dangling pointers that lead to a use-after-free condition. This can cause memory corruption or a crash, potentially allowing an attacker to execute arbitrary code or cause a denial of service.

ImageMagick is an open source software suite for displaying, converting, and editing raster image files. In ImageMagick versions prior to 7.1.2-7 and 6.9.13-32, an integer overflow vulnerability exists in the BMP decoder on 32-bit systems. The vulnerability occurs in coders/bmp.c when calculating the extent value by multiplying image columns by bits per pixel. On 32-bit systems with size_t of 4 bytes, a malicious BMP file with specific dimensions can cause this multiplication to overflow and wrap to zero. The overflow check added to address CVE-2025-57803 is placed after the overflow occurs, making it ineffective. A specially crafted 58-byte BMP file with width set to 536,870,912 and 32 bits per pixel can trigger this overflow, causing the bytes_per_line calculation to become zero. This vulnerability only affects 32-bit builds of ImageMagick where default resource limits for width, height, and area have been manually increased beyond their defaults. 64-bit systems with size_t of 8 b...

Squid is a caching proxy for the Web. In Squid versions prior to 7.2, a failure to redact HTTP authentication credentials in error handling allows information disclosure. The vulnerability allows a script to bypass browser security protections and learn the credentials a trusted client uses to authenticate. This potentially allows a remote client to identify security tokens or credentials used internally by a web application using Squid for backend load balancing. These attacks do not require Squid to be configured with HTTP authentication. The vulnerability is fixed in version 7.2. As a workaround, disable debug information in administrator mailto links generated by Squid by configuring squid.conf with email_err_data off.

vLLM is an inference and serving engine for large language models (LLMs). From versions 0.10.2 to before 0.11.1, a memory corruption vulnerability could lead to a crash (denial-of-service) and potentially remote code execution (RCE), exists in the Completions API endpoint. When processing user-supplied prompt embeddings, the endpoint loads serialized tensors using torch.load() without sufficient validation. Due to a change introduced in PyTorch 2.8.0, sparse tensor integrity checks are disabled by default. As a result, maliciously crafted tensors can bypass internal bounds checks and trigger an out-of-bounds memory write during the call to to_dense(). This memory corruption can crash vLLM and potentially lead to code execution on the server hosting vLLM. This issue has been patched in version 0.11.1.

Argo Workflows is an open source container-native workflow engine for orchestrating parallel jobs on Kubernetes. Argo Workflows versions prior to 3.6.12 and versions 3.7.0 through 3.7.2 expose artifact repository credentials in plaintext in workflow-controller pod logs. An attacker with permissions to read pod logs in a namespace running Argo Workflows can read the workflow-controller logs and obtain credentials to the artifact repository. Update to versions 3.6.12 or 3.7.3 to remediate the vulnerability. No known workarounds exist.

Argo Workflows is an open source container-native workflow engine for orchestrating parallel jobs on Kubernetes. Versions prior to 3.6.12 and versions 3.7.0 through 3.7.2 contain a Zip Slip path traversal vulnerability in artifact extraction. During artifact extraction the unpack/untar logic (workflow/executor/executor.go) uses filepath.Join(dest, filepath.Clean(header.Name)) without validating that header.Name stays within the intended extraction directory. A malicious archive entry can supply a traversal or absolute path that, after cleaning, overrides the destination directory and causes files to be written outside the /work/tmp extraction path and into system directories such as /etc inside the container. The vulnerability enables arbitrary file creation or overwrite in system configuration locations (for example /etc/passwd, /etc/hosts, /etc/crontab), which can lead to privilege escalation or persistence within the affected container. Update to 3.6.12 or 3.7.3 to remediate the ...

A vulnerability in the DocugamiReader class of the run-llama/llama_index repository, up to version 0.12.28, involves the use of MD5 hashing to generate IDs for document chunks. This approach leads to hash collisions when structurally distinct chunks contain identical text, resulting in one chunk overwriting another. This can cause loss of semantically or legally important document content, breakage of parent-child chunk hierarchies, and inaccurate or hallucinated responses in AI outputs. The issue is resolved in version 0.3.1.

A vulnerability in the ObsidianReader class of the run-llama/llama_index repository, specifically in version 0.12.27, allows for hardlink-based path traversal. This flaw permits attackers to bypass path restrictions and access sensitive system files, such as /etc/passwd, by exploiting hardlinks. The vulnerability arises from inadequate handling of hardlinks in the load_data() method, where the security checks fail to differentiate between real files and hardlinks. This issue is resolved in version 0.5.2.

A path traversal vulnerability exists in run-llama/llama_index versions 0.12.27 through 0.12.40, specifically within the `encode_image` function in `generic_utils.py`. This vulnerability allows an attacker to manipulate the `image_path` input to read arbitrary files on the server, including sensitive system files. The issue arises due to improper validation or sanitization of the file path, enabling path traversal sequences to access files outside the intended directory. The vulnerability is fixed in version 0.12.41.

The `SimpleDirectoryReader` component in `llama_index.core` version 0.12.23 suffers from uncontrolled memory consumption due to a resource management flaw. The vulnerability arises because the user-specified file limit (`num_files_limit`) is applied after all files in a directory are loaded into memory. This can lead to memory exhaustion and degraded performance, particularly in environments with limited resources. The issue is resolved in version 0.12.41.

A malicious user may submit a specially-crafted complex payload that otherwise meets the default request size limit which results in excessive memory and CPU consumption of Vault. This may lead to a timeout in Vault’s auditing subroutine, potentially resulting in the Vault server to become unresponsive. This vulnerability, CVE-2025-6203, is fixed in Vault Community Edition 1.20.3 and Vault Enterprise 1.20.3, 1.19.9, 1.18.14, and 1.16.25.