CYBER THREATCAST

Patch and inspect internet-facing WordPress, Joomla, Craft CMS, MaxSite CMS, MetInfo CMS, and Joomla JCE deployments now, prioritizing file-upload, RCE, SSRF, and deserialization flaws. Australia’s Cyber Security Centre warned that attackers are exploiting vulnerable CMS platforms and plugins at scale to deploy webshells for persistent remote access, and small and medium-sized Australian businesses have already been affected. Treat unpatched CMS instances as potentially exposed: review recently modified web files, remove unauthorized webshells, and verify plugin versions against available fixes.

Update Microsoft Defender’s Malware Protection Engine to version 1.1.26060.3008 or later across Windows 10 and Windows 11 systems. CVE-2026-50656 allows local privilege escalation to SYSTEM on fully patched Windows systems, and public proof-of-concept code was available before Microsoft issued the emergency update. Organizations should verify engine versions rather than relying on OS patch status alone, because the affected component is Microsoft Defender’s protection engine.

Reset FortiGate SSL-VPN administrative and VPN credentials, enforce MFA, patch FortiOS, and review logs for suspicious access; then inventory Ruckus and ASUS routers for known unpatched vulnerabilities. Cybercriminals are selling access to about 74,000 Fortinet FortiGate devices using apparently legitimate harvested credentials, and activity is still adding victims. Cisco Talos also reports China-nexus UAT-7810 exploiting known vulnerabilities in unpatched Ruckus and ASUS routers to expand Operational Relay Box infrastructure with custom malware.

Audit build systems and applications for malicious package versions: remove @injectivelabs/sdk-ts 1.20.21 and upgrade to the clean 1.20.23 release, and remove Braintree.Net versions 3.35.8 through 3.36.1 if present. Hackers compromised the Injective Labs SDK repository and published an npm wallet stealer aimed at private keys and seed phrases; Socket also reported NuGet packages impersonating PayPal Braintree’s .NET SDK that intercept live payment card data and exfiltrate merchant API keys and host secrets. Rotate any exposed wallet, payment, merchant, and host secrets touched by affected builds.

Disable autonomous spending, auto-mode, and auto-review execution in AI agents unless every tool action gets explicit human approval. Zscaler ThreatLabz documented active indirect prompt-injection campaigns using hidden webpage instructions, SEO poisoning, fake package content, metadata, and embedded content to make agents with financial capabilities send cryptocurrency without approval. Separate research showed prompt injection in repository files can manipulate Claude Code and OpenAI Codex review workflows into silent attacker-controlled code execution, with auto-mode or auto-review increasing the risk.