Out-of-bounds write in the php-src PHAR extension risks heap corruption and remote code execution
Attackers craft request URIs with a specific ending to trigger a logic flaw in zim_Phar_webPhar. The code incorrectly uses the full server-side path length instead of the base filename length when computing the offset, so the pointer lands past the end of the SAPI request_uri buffer. That causes a heap buffer over-read and a single-byte NUL out-of-bounds write. The memory corruption can overwrite critical data structures, crash the process, or enable remote arbitrary code execution, seriously threatening integrity and availability.
Component
php-src is the core source repository of the PHP interpreter. Its built-in PHAR extension provides JAR-like archive and stream handling for PHP. In non-CGI SAPI modes such as the CLI server, the webPhar function handles 301 redirect requests, parses URI paths, and maps them to PHAR files in nested local directories so static resources can be auto-distributed and routed.
Correct the length calculation so offsets use the proper base filename length, add strict bounds checks on pointer ranges to prevent buffer overflows, and disable unnecessary PHAR auto-loading in production to shrink the attack surface.
CSSA Exclusive Early Warning
10.0 Critical
Unauthenticated LibreTranslate metrics endpoint leaks API keys and enables full privilege abuse
Attackers can hit the unauthenticated /metrics endpoint and read raw API keys exposed in Prometheus metric labels. When recording request duration and active connections, the system embeds the caller's real API key as a label value, and /metrics is publicly reachable by default with no extra token. Any network-reachable attacker can collect every active user's credentials without authentication, fully impersonate those users, and may exhaust monitoring resources as label cardinality grows without bound.
Component
LibreTranslate is an open-source machine-translation service built as microservices and meant to give developers an easy local translation API. It accepts text over HTTP and returns translations, relying on backend model inference and front-end routing for real-time conversion across language pairs. Deployments aim to streamline multilingual content processing and scale via modular extensions for high-concurrency international workloads.
Type
Exposure of Sensitive Information to an Unauthorized Actor (CWE-200)
Disable public metrics collection, or require strict authentication on /metrics when it is enabled. Do not store plaintext secrets in monitoring labels—hash or redact them instead of showing raw keys. Regularly review logging and configuration policies, follow least privilege, and avoid over-exposing internal state.
CSSA Exclusive Early Warning
10.0 Critical
LibreTranslate client-script obfuscation grants API-key auth exemption, letting attackers bypass access control for unauthorized remote calls
Attackers abuse flawed identity checks based on the User-Agent header and rCookie on the /js/app.js route. Crafted HTTP requests recover the real API-key secret wrapped in JavaScript obfuscation. Because that crypto depends only on client-environment signals and can be reversed in a lightweight JS sandbox without a full browser DOM, attackers easily restore valid credentials and then call /translate indefinitely without a key, fully bypassing rate limits and billing and exhausting service resources while damaging business-logic integrity.
Component
LibreTranslate is an open-source local machine-translation engine that gives applications plug-and-play multilingual translation. Its modular backend and front-end interfaces support RESTful API integration, covering real-time text translation, source-language detection, and large-scale concurrent requests, with the goal of a decentralized, privacy-preserving translation infrastructure.
Stop trusting client-controlled signals such as User-Agent or Cookie on the server; enforce access control based on server-side session state or strong authentication protocols, and keep sensitive configuration out of front-end code.
CVE Intelligence
2
CVE-2026-120276
CVSS
10.0 Critical
2026-11-02
Command injection in the WAVLINK WN536AX6-A web management interface enables remote code execution
In WAVLINK WN536AX6-A firmware V260507_2.0.0, function sub_416760 in /bin/ioos of the Web Management Interface has a command-injection flaw (CWE-77). By manipulating the group_name parameter, an attacker can inject OS commands and run arbitrary commands remotely. The flaw allows unauthenticated remote attack, and exploit code is public, making automated abuse easy. All WN536AX6-A devices on V260507_2.0.0 are affected. A crafted network request alone can trigger the issue with no user interaction and fully control the device.
Component
WAVLINK WN536AX6-A is a wireless router whose Web Management Interface is used for device configuration and administration.
Risks
Complete system control: Attackers can run arbitrary system commands on the router, fully control the device, change configuration, steal data, or plant backdoors
Remote no-interaction attack: Attackers can strike over the network with no victim clicks or interaction
Public exploit code: Because a PoC is public, the flaw faces high risk of large-scale automated scanning and attack
In Prefab UI 0.14.0 through 0.20.2, the Generative UI sandbox has a critical permission-misconfiguration flaw. The sandbox is meant to isolate untrusted code from large language models (LLMs), but it starts the Deno subprocess with unrestricted --allow-read, --allow-write, and --allow-net and exposes Deno APIs on the global JavaScript object. Because Pyodide's js module proxies that global object, attacker-injected Python can call Deno.readTextFileSync and Deno.writeTextFileSync directly. Attackers can read and write arbitrary files as the server OS user and persist code execution by writing .pth files into the server virtualenv's site-packages. A common remote path is FastMCP's GenerativeUI provider, whose generate_prefab_ui tool forwards the code parameter straight into the sandbox. On unauthenticated FastMCP servers any reachable client can exploit this; with authentication, any client authorized to call the tool can. Fixed in Prefab UI 0.21.0.
Component
Prefab UI is an open-source component library for generating user interfaces. Its Generative UI feature runs caller-supplied Python in a sandbox (Pyodide and Deno), often for LLM-generated UI code.
Risks
Complete system control: Attackers can run arbitrary code on the victim system and, depending on privileges, install programs, view/change/delete data, or create fully privileged accounts
File I/O and persistence: Attackers can read/write arbitrary files as the server OS user and persist backdoors via .pth files
No-user-interaction attack: Through FastMCP, attackers can trigger the flaw remotely without user interaction—especially severe when authentication is not configured