Noma Labs found a critical flaw in Ruflo, an open-source platform for orchestrating swarms of AI agents that acts as a harness for coding agents like Claude Code and Codex. Tracked as CVE-2026-59726 and scored 10.0, the flaw is that Ruflo's default Docker deployment exposed its Model Context Protocol bridge to the network with no authentication, offering 233 callable tools. A single unauthenticated request could invoke a shell command tool to run code in the bridge container, read AI provider API keys, steal stored conversations, and poison the platform's persistent agent memory, which can steer future outputs even after the attacker leaves. The flaw is in Ruflo, not in the coding agents.
Security firms ThreatBook and Imperva report attackers targeting a critical flaw in Fastjson, Alibaba's widely used JSON library for Java. On affected Spring Boot applications, a malicious JSON request can execute code without authentication, running with the privileges of the Java process. Tracked as CVE-2026-16723 with an Alibaba-assigned score of 9.0, the confirmed chain needs Fastjson 1.2.68 through 1.2.83, a Spring Boot executable fat-JAR, a network-reachable path that feeds attacker-controlled JSON to the parser, and SafeMode left at its disabled default. Notably, the AutoType feature can stay off and no classpath gadget is required. As of July 25, Alibaba had not shipped a fixed 1.x version.
Cl0p affiliates are exploiting internet-exposed PTC Windchill and FlexPLM product lifecycle platforms through an unauthenticated remote code execution flaw, in a data theft extortion campaign. Product lifecycle management systems track a product from design to retirement and hold computer-aided design files, bills of materials, engineering data, and workflows, making them a concentrated store of intellectual property for manufacturers in automotive, aerospace, defense, medical, and electronics. Cl0p's established pattern is mass exploitation of one enterprise product followed by extortion over stolen data rather than encryption. PTC patched a critical deserialization flaw in the Windchill data management component earlier this year that has already seen exploitation.
Researchers disclosed a critical unauthenticated remote code execution flaw in Hugging Face's LeRobot, the open-source framework used to train and deploy ML models on physical robots. CVE-2026-25874 sits in the framework's web interface, which by default listens on all network interfaces with no authentication - quick for demos, but a hard fail when the demo box ends up on a corporate network. There is no patch yet. Hugging Face has been notified but hasn't released a fix. Particularly serious because LeRobot is usually attached to actual robotic hardware, so a compromise can mean unsafe physical actions.
Wordfence has seen more than 170 live exploit attempts against CVE-2026-3844, a critical unauthenticated arbitrary file upload in the Breeze Cache WordPress plugin from Cloudways. Breeze has roughly 400,000 active installations, making this one of the larger exposure events of the month. The flaw lives in the fetch_gravatar_from_remote function, which fetches avatar images from an arbitrary remote URL and saves them locally without validating the downloaded file's MIME type - so an attacker can point it at a .php payload and drop a webshell directly into a web-accessible directory. The attack is only possible when the 'Host Files Locally - Gravatars' add-on is enabled, which is not the default, but any site that turned it on for performance reasons is wide open. Cloudways shipped the fix as Breeze 2.4.5 earlier this week; as of publication only about 138,000 of the 400,000 installations had downloaded the patched version, leaving hundreds of thousands of sites exposed to a pre-auth RCE with 9.8 CVSS.
TrustStrike Labs