NLTK versions 3.10.0 through 3.10.3 contain a compile-time regular-expression denial of service (ReDoS) vulnerability in their regex search and tokenizer functions. This vulnerability arises because the third-party regex engine used by NLTK compiles patterns with many capturing groups in roughly quadratic time, allowing an attacker to supply a crafted regex pattern that causes excessive CPU consumption during compilation. This leads to denial of service by exhausting CPU resources. The vulnerability does not affect confidentiality or integrity and is remediated in NLTK version 3.10.4.
NLTK versions prior to 3.10.4 contain a vulnerability in four hypernym-walking methods of the WordNet Synset class that perform recursive graph traversal without cycle detection or depth limits. An attacker who can supply a malicious WordNet corpus file with cyclic or excessively deep hypernym pointer graphs can cause these methods to recurse indefinitely, resulting in a RecursionError that aborts the traversal and potentially crashes the application process. This vulnerability impacts availability only and requires no special privileges beyond loading attacker-controlled corpus data. The issue is fixed in NLTK version 3.10.4.
NLTK's SeekableUnicodeStreamReader.readline function has a performance issue where reading a single very long line without a newline character causes quadratic CPU usage due to repeated rescanning of an accumulating buffer. An attacker controlling the corpus file content can exploit this by providing a large unterminated line, causing significant CPU consumption and denial of service. The issue affects NLTK versions prior to 3.10.4 and has been fixed in version 3.10.4.
NLTK's XMLCorpusView.read_block method suffers from a quadratic CPU exhaustion vulnerability when processing deeply nested XML corpus tags. This occurs because the method rebuilds the full root-to-node context path on every start tag while searching for a matching element, causing O(n^2) computational complexity relative to the nesting depth. An attacker who can supply or influence the XML corpus file can cause high CPU usage, leading to denial of service. The vulnerability affects versions prior to 3.10.4 and has no impact on confidentiality or integrity, only availability.
NLTK versions 3.10.0 through 3.10.3 contain a vulnerability in the StreamBackedCorpusView constructor where the os.stat call on a file path string leaks metadata about files outside the allowed sandbox roots. This bypasses the pathsec enforcement intended to restrict filesystem access, exposing file existence, exact size, and modification time without disclosing file content. The vulnerability requires an attacker to influence the file path string passed to the constructor. The issue is fixed in version 3.10.4.
NLTK versions up to 3.10.3 have a vulnerability in the prover9 and mace4 configuration that allows local code execution when a relative binary_location is used. The issue arises because the code accepts relative paths for the prover9/mace4 binaries without validating that they are absolute paths, leading to execution of attacker-planted binaries in the current working directory. This flaw is due to missing path validation that other similar components in NLTK enforce. A fix is available in version 3.10.4, which rejects relative binary locations and enforces absolute-only paths.
The ct-fivem npm package contains malicious code that fully compromises any computer on which it is installed or running. This compromise includes potential unauthorized access and control, necessitating immediate rotation of all secrets and keys from a separate, secure machine. Removal of the package alone does not guarantee elimination of all malicious software introduced by its installation.
Multiple memory-safety vulnerabilities in Ruby's regular expression engine and string case conversion affect versions prior to 4.0.10, 3.4.13, and 3.3.14. These flaws could allow remote unauthenticated attackers to cause crashes, read memory, or execute arbitrary code when processing crafted input, particularly in applications using Rack-based frameworks like Rails and Sinatra.
CVE-2026-133327 is a code injection vulnerability in Dolibarr ERP CRM versions 23.0.0 through 23.0.4. It affects the dol_convert_file function in the Document Upload component, allowing remote attackers to inject code. A public exploit exists. Upgrading to version 24.0.0 mitigates the issue.
CVE-2026-122676 is a Blind SQL Injection vulnerability in the PublishPress Authors plugin affecting versions prior to 4.20.0. This flaw allows an attacker with low privileges to execute crafted SQL commands that can expose sensitive data. The vulnerability has a high severity score of 8.5 and impacts confidentiality and availability. A fix is available in version 4.20.0.
The npm package sirens-lamentt version 99.0.2 contains malicious code that executes a postinstall script. This script sends an HTTPS GET request to a hardcoded webhook.site URL, transmitting the installer's IP address and install event data to a third-party collector. The behavior is consistent with a dependency confusion attack probe targeting internal package names.
Guard Core versions prior to 3.13.0 contain a vulnerability where the X-Forwarded-For (XFF) header is extracted without IP address validation and used directly in Redis key names and KEYS glob patterns. This allows attacker-controlled strings to cause cross-client counter contamination, unbounded Redis key growth, blocking Redis scans, and potential server errors due to unhandled exceptions. The issue affects deployments where the extracted XFF entry is attacker-controllable and trusted proxies are configured. No data exposure is reported. A fix is available in version 3.13.0 that validates the IP and escapes glob metacharacters or replaces the KEYS command usage.
Guard-core versions prior to 3.13.0 in enhanced detection mode preprocess input by collapsing all whitespace, including CR and LF characters, before pattern matching. This preprocessing removes the CRLF characters required by two built-in detectors (http_split and cmd_injection), causing them to miss payloads that include CRLF sequences. As a result, HTTP response splitting and newline-anchored command injection payloads evade detection, although other attack types remain detected. The issue is fixed in version 3.13.0 by preserving CRLF during whitespace collapse or adjusting detection to run on preprocessed content.
A quadratic regular-expression denial of service (ReDoS) vulnerability exists in guard-core's built-in template-injection detection pattern starting from version 3.5.0 up to but not including 3.13.0. This flaw allows unauthenticated remote clients to freeze the asyncio event loop of applications embedding guard-core by sending specially crafted requests containing repeated '{' characters. The vulnerable pattern executes without timeout protection, causing the event loop to block for seconds per request without triggering detection or rate limiting. This regression was introduced in version 3.5.0, undoing previous protections. The issue is fixed in version 3.13.0.
k8s-mcp-server versions prior to 1.1.0 running in default SSE transport mode expose an unauthenticated HTTP endpoint on port 8080 bound to all interfaces. This endpoint allows any network-accessible client to enumerate and invoke Kubernetes tools, including destructive operations, without authentication. The server does not enforce authorization or origin checks, and the default configuration permits state-changing commands. This vulnerability is due to missing authentication on critical functions and is classified as CWE-306.
CodeWhale versions 0.8.67 and earlier contain a vulnerability in the interactive Agent profile where certain shell commands are auto-approved without user prompt. This occurs because commands classified as "parallel read-only" such as 'fd', 'rg', and 'git grep' bypass the approval prompt despite having flags that enable arbitrary external program execution. The flaw allows silent remote code execution on the developer host by executing attacker-controlled commands without explicit approval. The issue is fixed in version 0.9.0.
Multiple security issues affecting the nginx HTTP/2 implementation were disclosed in November 2026. AWS acknowledges these issues and recommends upgrading to nginx versions 1.30.7 (stable) or 1.31.9 (mainline) where the vulnerabilities have been addressed. AWS is also deploying mitigations globally. These vulnerabilities occur at the HTTP/2 protocol layer before a full HTTP request is formed, limiting the effectiveness of AWS WAF rules in detecting or blocking exploitation attempts.
CodeWhale versions 0.8.67 and earlier have a vulnerability where the automation subsystem's auto_approve setting defaults to true, bypassing task-path hardening fixes. This allows a single approved automation_create call to create a durable, recurring, fully auto-approved agent that runs mutating tools without human approval. If allow_shell is enabled, this leads to recurring silent remote code execution. The issue is fixed in version 0.9.0.
A regular expression denial of service (ReDoS) vulnerability exists in eml-parser versions prior to 3.0.4 in the Received-header parsing function. The vulnerability arises from a dynamically constructed regex with multiple greedy capture groups followed by a normalized date literal that may not match the original header line, causing catastrophic backtracking with O(n^4) complexity. This leads to high CPU usage and can hang parser workers on crafted email headers. The issue is fixed in version 3.0.4.
A vulnerability in containerd's image manifest processing causes the containerd daemon to crash when handling images with a malformed digest in a subject-annotated descriptor. This can disrupt container operations, especially in Kubernetes environments where repeated image pull retries can repeatedly crash containerd on worker nodes. The issue is fixed in containerd versions 2.2.10, 2.3.7, and 2.4.2. Users should update to these versions or restrict image pulls to trusted registries as a workaround.