Critical Watch

FFmpeg is embedded in countless media-processing pipelines, servers, and apps, and this patched RTSP redirect-loop flaw (with public POC) lets a malicious RTSP server pin a CPU core indefinitely, posing a realistic DoS risk to any service that ingests untrusted streams.

A second ubiquitous FFmpeg vulnerability where a crafted HLS playlist triggers an infinite loop in the HLS demuxer, letting remote attackers exhaust CPU and I/O on any host transcoding attacker-supplied media - a direct availability threat to video/streaming backends.

Excelize is a widely used Go library for spreadsheet handling, and this patched flaw (flagged with threat-intel) lets an attacker-controlled .xlsx pin a CPU core when an app calls GetRows/Rows.Next, making any Go service that parses user-uploaded spreadsheets vulnerable to easy DoS.

Intego Antivirus for Windows contains a link-following (junction) flaw letting any local unprivileged user escalate to SYSTEM via its optimization module, turning a defensive security product into a reliable local privilege-escalation foothold on endpoints that deploy it.

Iskratel Innbox GPON ONT devices are mass-deployed by ISPs, and this unauthenticated OS command injection in the login.xgi endpoint grants root to anyone who can reach the web interface, enabling large-scale device takeover and botnet recruitment of subscriber equipment.

vLLM is one of the most widely deployed LLM inference engines in production AI stacks, and this flaw lets a single crafted multimodal request crash the shared serving process, denying service to every tenant on the engine.

Payload CMS is a popular headless CMS, and this access-control bypass lets any authenticated user overwrite another account's password despite field-level update restrictions, enabling straightforward account takeover of privileged users.

The WP Coder WordPress plugin flaw lets an Editor-level user save and execute arbitrary PHP, giving full server/site compromise, and it already has public exploit code and known-actor activity making it a near-term mass-exploitation risk for WordPress fleets.

Envira Gallery is a widely installed WordPress gallery plugin, and this actively-exploited stored XSS lets an Author-level user plant persistent script that runs in admin/visitor browsers, a common foothold for site defacement and admin session hijacking.

Optimole is deployed on hundreds of thousands of WordPress sites, and this flaw exposes stored third-party image-service credentials to any authenticated user down to the Subscriber role, enabling low-effort credential theft and lateral abuse of the connected service.

NextChat (ChatGPTNextWeb) is a massively deployed self-hosted AI chat frontend, and this unauthenticated SSRF in its /api/proxy endpoint-backed by public exploit code and known-actor threat intel-lets attackers pivot into internal networks and cloud metadata services.

SciPhi R2R is a widely used RAG/AI framework, and the shipped default JWT signing secrets let unauthenticated attackers forge valid tokens and fully bypass authentication on any deployment that hasn't rotated the keys, with threat-intel activity already observed.

Payload CMS is a popular headless CMS, and this catastrophic-backtracking ReDoS in its multipart Content-Type validation-with a patch, public exploit, and known-actor intel-lets unauthenticated attackers trivially hang servers and cause denial of service across hosted apps.

Docling is a broadly adopted document-processing library in AI/RAG pipelines, and this patched local file disclosure (threat-intel confirmed) lets an attacker-controlled HTML document exfiltrate sensitive local files via file:// references during rendering.

The Razorpay for WooCommerce plugin runs on a large base of e-commerce stores, and this unauthenticated authorization flaw lets attackers tamper with shipping details on orders they don't own, enabling fraud and order redirection.

Unlimited Elements for Elementor is installed on well over a million WordPress sites, and this path traversal lets a privileged (or Editor, when a common setting is on) user plant executable files for code execution, with known threat actors already tracked against it.

Mailchimp for WooCommerce runs on a very large number of e-commerce sites, and this unauthenticated flaw lets attackers modify or delete other customers' abandoned-cart records with no login, with active threat-actor interest making it a near-term exploitation risk.

A CSRF in the CoCart WooCommerce plugin (before 4.9.7) lets an unauthenticated attacker create a rogue administrator account by luring a logged-in admin to a malicious page, delivering full site takeover at CVSS 8.8.

MetForm is a widely deployed Elementor form builder, and this flaw exposes all stored form submissions through its REST API to any unauthenticated visitor, leaking PII at scale with known actors already probing it.

WP Ultimate Review before 2.4.4 lets unauthenticated visitors persistently break front-end page rendering on sites with review submission enabled, a denial-of-service that takes pages offline and is already associated with known malicious activity.

Loco Translate is one of the most widely installed WordPress translation plugins (well over a million active sites), and this path-traversal flaw has a public exploit letting translator-level users read arbitrary files outside the web root, exposing wp-config.php secrets.

The Model Context Protocol reference Fetch Tool (mcp-server-fetch) is embedded across a rapidly growing base of AI agent deployments, and this unauthenticated SSRF lets remote callers pivot fetch_url into internal cloud metadata and other restricted endpoints.

Unlimited Elements for Elementor is a popular add-on on hundreds of thousands of WordPress sites, and this blind SQL injection with a public PoC lets an authenticated attacker read or tamper with the site database.

The same mass-deployed Loco Translate plugin also has a public-exploit stored XSS that fires in the admin panel, giving a low-privileged translator a path to admin session/account takeover.

InternLM MindSearch carries an unauthenticated critical (CVSS 9.3) code-injection RCE with a public exploit in its Planner Agent, so any internet-exposed instance can be fully compromised remotely.