Threat Brief — 2026-07-19 — Patch Windows, WordPress, and 7-Zip Now

Executive summary: Three critical RCE vulnerabilities demand immediate patching — 7-Zip, WordPress Core, and the already-KEV'd SharePoint zero-day continues to be exploited. Meanwhile, Microsoft warns of a surge in ACR Stealer attacks against enterprise customers, and two APT campaigns (ToddyCat and Armored Likho) are actively compromising corporate email and stealing data using novel toolsets. Device Code Phishing via Microsoft OAuth is gaining traction as a credential-theft vector that bypasses traditional URL-checking controls.

Top items

• 7-Zip RCE patched (v26.02) — A remote code execution flaw in the world's most widely used archive tool can be triggered simply by opening a crafted compressed file. Update all endpoints immediately; the attack surface is enormous given 7-Zip's ubiquity. Source

• WordPress Core "wp2shell" RCE — public exploits released — Critical remote code execution flaws in WordPress Core now have working public exploits, making unpatched sites trivially compromisable. Any internet-facing WordPress instance is at immediate risk of full server takeover. Source

• ACR Stealer surge against enterprise — Microsoft reports a significant uptick in ACR Stealer attacks harvesting browser-stored passwords, OAuth tokens, and sensitive documents from enterprise customers. This credential-extraction capability directly enables downstream lateral movement. Source

• ToddyCat "Umbrij" — Gmail OAuth compromise tool — APT group ToddyCat has deployed a new tool specifically designed to hijack corporate Gmail via OAuth authorization tokens, granting persistent access to email without needing credentials. This extends ToddyCat's espionage toolkit beyond their known loader/backdoor arsenal. Source

• Armored Likho APT — BusySnake Stealer campaign — Active targeted phishing campaign hitting organizations in Russia, Kazakhstan, and Brazil using AI-generated loaders and a new Python-based "BusySnake Stealer." The use of AI to generate malware loaders signals continued weaponisation of generative AI by established APT groups. Source

• Microsoft Device Code Phishing — Attackers are abusing the OAuth 2.0 Device Authorization Grant (designed for IoT/Smart TV authentication) to phish Microsoft credentials. This bypasses URL-based security checks because the phishing flow runs through legitimate Microsoft login.microsoftonline.com endpoints. Conditional access policies should be reviewed to restrict device-code flows. Source

• Schneider Electric license manager — CVE-2024-2658 — A vulnerability in FlexNet Publisher used by Schneider Electric's license manager could enable SYSTEM-level privilege escalation on industrial systems. Given FlexNet's deployment across numerous ICS environments, exposure may extend beyond Schneider's products. Source

• ScreenConnect delivering AsyncRAT at scale — A large malicious infrastructure campaign is distributing AsyncRAT trojans through disguised ScreenConnect (ConnectWise) remote-access tool bundles. The campaign's C2 infrastructure has been mapped; organisations should block identified IOCs and audit ScreenConnect deployments. Source

• CoreDNS DoS vulnerabilities (CVE-2026-62309, CVE-2026-62299) — Two remote denial-of-service flaws in CoreDNS plugins (proxyproto and rewrite/EDNS0) can be triggered by crafted single packets. Any deployment using these plugins in production DNS infrastructure should patch immediately to prevent service disruption. Sources: [1, 2]

• Squid proxy — memory disclosure and corruption (CVE-2026-47729, CVE-2026-50012) — Two vulnerabilities in Squid's FTP gateway and cache_digest handling could leak memory or corrupt memory remotely. Organisations running Squid as a forward/reverse proxy should patch. Sources: [1, 2]

• SharkNinja robot vacuum remote control — A cloud-platform vulnerability allows remote internet-based control of Shark robot vacuums, and can leak stored Wi-Fi credentials. IoT device inventories should be reviewed for cloud-connected consumer appliances on corporate networks. Source

• OpenClaw agent — malicious skills and data exfiltration risks — Security analysis of the OpenClaw AI agent reveals exploitable vulnerabilities, malicious skill injection, and data leakage paths. Organisations deploying AI agents should review skill provenance and enforce sandboxing. Source

Themes

OAuth and token abuse across the board — ToddyCat's Umbrij tool, Microsoft Device Code Phishing, and ACR Stealer all target OAuth tokens and browser-stored credentials rather than passwords. The shift from credential theft to token hijacking means traditional MFA alone is insufficient; conditional access and token lifecycle controls are essential.

AI weaponisation maturing — Armored Likho's AI-generated loaders, OpenClaw's exploitable agent architecture, and the demonstrated $100 AI-model poisoning experiment all point to adversarial AI moving from theory to operational deployment.

ICS supply-chain exposure — Schneider Electric's FlexNet vulnerability and the ViPNet HelloNet campaign both demonstrate that trust relationships in industrial software supply chains remain a persistent gap.