ONNX versions through 1.21.0 have a vulnerability in shape inference for RNN, GRU, and LSTM nodes from opsets 1–6 where an out-of-bounds read occurs if the first input's rank is not three. This can cause a process using onnx.shape_inference.infer_shapes() to crash, resulting in denial of service. The issue is fixed in ONNX 1.22.0 by validating the input rank before accessing dimensions.
vLLM deployments serving DeepSeek-OCR family models are vulnerable to a denial of service caused by a crafted PNG image. A legitimate half-black/half-white 1024×1024 image triggers a zero-sum pixel value condition that causes the model's input validation to return None. This leads to an assertion failure in the engine worker process, crashing the entire service and requiring manual restart. The issue affects versions from 0.11.1 up to but not including 0.32.0 and is fixed in version 0.32.0.
vLLM versions prior to 0.32.0 contain a vulnerability where an out-of-vocabulary token ID equal to the vocabulary size can bypass input validation and cause a CUDA device-side assert, crashing the engine. This affects models whose vocabulary size is an exact multiple of the embedding padding size (64), such as Llama-2-family models with a vocab size of 32000. The issue occurs specifically when using a single GPU (tensor parallel size 1) configuration. The flaw was fixed in vLLM version 0.32.0 by correcting the maximum accepted token ID validation.
A denial of service vulnerability exists in github.com/go-jose/go-jose/v4 where parsing deeply nested JSON structures can cause the Go runtime to hit its stack limit and crash the application. This affects versions prior to 4.1.5 and up to 3.0.5. The issue is fixed in version 4.1.5. Applications parsing untrusted JSON with affected versions are exposed to crashes.
A denial of service vulnerability exists in github.com/go-jose/go-jose/v4 when decrypting ciphertexts using the CBC-HMAC content encryption algorithms A128CBC-HS256, A192CBC-HS384, or A256CBC-HS512. Insufficient validation of parameters can cause a panic due to indexing into an empty buffer on untrusted input. This affects versions prior to 4.1.5 and up to 3.0.5. A fix is available in version 4.1.5.
The graphql-scalars package's PhoneNumber scalar is vulnerable to a regular expression denial of service (ReDoS) due to an exponential backtracking flaw introduced in version 1.24.1. This flaw allows specially crafted inputs of digits followed by a failing character to stall the JavaScript thread for seconds, with processing time doubling for each additional digit. The vulnerability affects all versions from 1.24.1 up to and including 2.0.0. The regex runs during GraphQL document validation and variable coercion, before any authentication or authorization, enabling unauthenticated requests to cause denial of service. A fix is available in version 2.0.1.
LangGraph API versions from 0.1.2 up to but not including 0.15.0 that use JavaScript or TypeScript custom authentication with namespace rewriting for Store authorization may allow authenticated users to access or modify data outside their authorized namespace. This vulnerability does not affect Python custom authentication or handlers that reject unauthorized namespaces. The issue is fixed in version 0.15.0.
filelock versions 4.1.0 through 4.2.0 contain a vulnerability in SoftReadWriteLock where a local user with write access to the lock's parent directory can replace the lock directory tree with symbolic links after the first lock acquire. This causes the lock user process to write lock records through the symbolic links, potentially corrupting files such as a git repository's HEAD file. The vulnerability requires local access and specific directory conditions but does not expose confidentiality or availability impacts. It has a low CVSS score of 3.3 and is fixed in version 4.2.1.
The filelock library versions 3.30.0 through 4.2.0 have a vulnerability in StrictSoftFileLock and AsyncStrictSoftFileLock where claim directories are created without applying the caller's restrictive mode, resulting in directories with permissions influenced by the process umask or parent directory ACLs. This can allow another local user with write access to these directories to unlink claim files, causing multiple holders to believe they hold the lock simultaneously, leading to integrity issues. The vulnerability requires a local attacker with low privileges and specific environment conditions such as a permissive umask or default ACLs. The issue is fixed in version 4.2.1.
filelock versions 3.30.0 through 4.2.0 have a vulnerability where StrictSoftFileLock and SoftReadWriteLock classes adopt a protocol directory created by another local user if it exists before the lock user's first acquire. This allows the other user to manipulate lock state files, causing integrity and availability issues such as multiple writers entering critical sections simultaneously and denial of service via spinning locks or memory exhaustion. The vulnerability requires local access and the ability to create directories in shared sticky-bit directories like /tmp before the lock user. It has no confidentiality impact but high integrity and availability impact. A fix is available in version 4.2.1.
The filelock library on Windows ignores the 'mode' argument when creating lock files, causing files intended to have restrictive permissions (0o600) to inherit the directory's ACL instead. This allows other local accounts with read or modify access to the directory to read or write the lock files and interfere with the locking mechanism. The issue affects filelock versions 3.10.0 through 4.2.0 on Windows and is confirmed by the maintainers. Unix systems are not affected. A fix is available in version 4.2.1.
filelock versions 3.20.4 through 4.2.0 contain a vulnerability in the ReadWriteLock and AsyncReadWriteLock mechanisms where a crafted SQLite lock database file can cause every acquire operation to stall or hang indefinitely, ignoring timeout and blocking settings. This is due to the lock code running PRAGMA journal_mode=MEMORY on a fresh SQLite connection for each acquire, which loads the entire schema from the file at the lock path. A local attacker who can place or overwrite this file can cause denial of service by making acquires hang or spin at 100% CPU. The impact is availability only; no confidentiality or integrity is affected. The vulnerability is fixed in filelock 4.2.1.
A vulnerability in filelock versions 3.22.0 through 4.2.0 allows a local user with write access to a lock marker file to forge the process ID (PID) recorded in the marker. This causes another process, running as the lock owner or root, to mistakenly consider the lock stale and remove the live marker, resulting in two processes concurrently holding the lock. The issue arises when the marker file is group- or other-writable due to explicit mode settings or permissive umask values in sticky or non-removable directories. The vulnerability has a medium severity with a CVSS score of 4.7 and impacts integrity but not confidentiality or availability. A fix is available in version 4.2.1.
A symlink swap vulnerability in filelock's SoftFileLock writability check allows a local user with directory write access to leak one bit of information about whether a path exists and is writable by the lock user. This occurs due to a time-of-check to time-of-use (TOCTOU) race condition involving os.lstat and os.access calls. The vulnerability affects filelock versions 3.32.6 through 4.2.0 and is mitigated by Linux kernel's protected symlinks feature in common sticky world-writable directories like /tmp. The impact is limited to leaking a single bit of path existence and writability information with high attack complexity and no integrity or availability impact. A fix is available in version 4.2.1.
The filelock library's UnixFileLock backend on Unix systems truncates and changes permissions of any regular file that another local user renames or hard-links onto the lock path. This occurs when the lock user acquires the lock and the file exists at the lock path, leading to unintended truncation and potential permission escalation. The vulnerability affects all filelock versions up to and including 4.2.0 and requires local user access with certain directory permission conditions. The issue is fixed in version 4.2.1.
A race condition vulnerability in filelock versions 3.30.0 through 4.2.0 allows a local unprivileged user to swap a symbolic link at the <path>.filelock directory after an lstat check but before subsequent operations. This can cause the lock user's process to create claim files in an attacker-chosen directory, potentially disrupting other locks by causing stranded claims that lead to denial of service until manually force-broken. The vulnerability requires write access to the directory containing <path>.filelock and is mitigated by patch 4.2.1.
filelock versions 4.0.0 through 4.2.0 contain a vulnerability in SoftReadWriteLock where the holder record refresh writes through a hard link or hangs on a named pipe (FIFO) placed by another local user who owns the lock's .rw directory. This can cause the lock user's write_lock or read_lock operations to hang indefinitely and the heartbeat thread to stop without notification, allowing multiple processes to believe they hold the lock simultaneously. The issue requires local user access and specific conditions such as permissive hard-link policies or ownership of the lock directory. The vulnerability has medium severity with no confidentiality impact but can cause integrity loss and high availability impact. A fix is available in version 4.2.1.
OpenTelemetry Java's W3C trace-context propagator has a vulnerability that allows crafted tracestate headers to cause excessive CPU consumption due to quadratic regex backtracking. This can lead to denial of service by blocking the parsing thread. The issue affects versions prior to 1.67.0 and is fixed in version 1.67.0. Mitigation includes upgrading to 1.67.0 or later or rejecting oversized tracestate headers before extraction.
LangGraph Python SDK v3 streaming client versions 0.4.0 through 0.4.5 improperly handle unencoded thread_id and assistant_id values in request paths, allowing relative path segments to redirect authenticated GET requests to other paths on the same server. This affects also langgraph versions 1.2.3 through 1.2.13 when used with vulnerable SDK versions. The vulnerability allows an attacker controlling these identifiers to cause the SDK to fetch and return data from arbitrary readable paths on the server, but does not allow modification of data. The issue is fixed in langgraph-sdk 0.4.6 and langgraph 1.2.14.
A vulnerability in the @graphql-tools/executor-legacy-ws package causes per-request WebSocket connection parameters to persist across different users. This stale authentication state leads to later users inheriting previous users' authentication credentials in the connection_init payload, potentially allowing unauthorized access to downstream GraphQL subscriptions. The issue affects versions prior to 1.1.36 and is fixed in version 1.1.36.