JFrog Security Real Time Posts
Last Updated On 18 Nov, 2026

JFrog Security Research is tracking CVE-2026-123007 and CVE-2026-123633, two CRITICAL-severity NGINX vulnerabilities (Nginx2Shell) that allow unauthenticated remote code execution in NGINX itself, independent of the application running behind it. We break down who is affected, the bug flow, the shipped fix, and how to detect exploitation attempts.

The JFrog Security Research team analyzed the supply chain hijack of Microsoft's agent-framework package on PyPI. Compromised versions 1.26.1 and 1.26.2 use Python .pth startup hooks to harvest Azure identities, CI/CD tokens, and AI developer secrets on every Python launch. If you installed either version, assume your environment is compromised.

A critical remote code execution vulnerability affecting OpenJDK Java 17 has been reported under CVE-2026-114938.

JFrog Security Research traced a fake interview operation run by the fictitious company AccuraSoftHire: a careers site, job ads on real boards, ID-verified LinkedIn recruiters, a poisoned take-home assignment and a 'camera driver' serving a malware loader to job candidates.

A one-letter typo, Content-Xength, is accepted as Content-Length by the Eclipse Vert.x HTTP/1.x decoder (CVE-2026-102479). We chain it into response-queue desync and cross-user session fixation behind real reverse proxies.

JFrog Security Research linked 42 armed GitHub repositories in the testmetest campaign to a hidden npm publisher, an active encrypted data sink and previously reported GitHub Actions cPanel exploitation infrastructure.

CVE-2026-111497 ("Brotli2Shell") is an out-of-bounds write in Google's Brotli decoder that reaches far beyond Node.js and HTTP. JFrog Security Research reproduced a remote crash in ClickHouse, native memory corruption in FFmpeg through JPEG XL, a .NET runtime crash, and an infinite loop in Apache Arrow, and released an open-source scanner that finds Brotli decoding paths version-only scanners miss.

JFrog Security Research is tracking CVE-2026-109618 and CVE-2026-109619, two CRITICAL-severity PHP core vulnerabilities (PHP2Shell) that turn unauthenticated requests into remote code execution and Heartbleed-style memory disclosure. We break down the bug flow, the shipped fix, and the fake PoCs that surfaced around the disclosure.

LiteLLM has been hijacked again: malicious `litellm` 1.107.0 wheels contain an interpreter-startup credential stealer and an hourly remote-code loader absent from the source distribution.

JFrog Security Research is tracking CVE-2026-92392, a HIGH-severity curl/libcurl use-after-free in HTTP/2 uploads that a malicious HTTPS server can turn into client-side remote code execution. We break down the bug, the published exploit, and the narrow conditions it needs.