ConnectWise suspended three services after attackers used a compromised Cloud Backup credential to access customers’ Microsoft 365 environments. Huntress observed unauthorized sign-ins in 162 tenants, and ConnectWise gave no timeline for restoring service.
Compromised suppliers and official distribution channels connect several other breaches: attackers put secret-stealing malware in OVHcloud’s GitHub releases and backdoors in some Rank Math PRO plugin packages. Rank Math’s clean update does not remove existing infections, leaving affected sites compromised even after an upgrade.
AI tools are both attack targets and tools for breaking in. GreyNoise reported that attackers used Codex and DeepSeek to compromise at least 440 PaperCut servers, while a patched Claude Code Desktop flaw let malicious HTML steal developers’ local credentials and private keys.
Microsoft is extending Project Perception to manage security risks from AI agents, adding discovery support for 35 local agent types to bring the total to 60, including Claude Code, OpenAI Codex and GitHub Copilot. The company also introduced Dynamic Permissions to limit agent access by task and duration, alongside Dynamic Agent Risk and Dynamic Conditional Access for agents.
More than 40,000 organizations received fake calendar invitations designed to steal Microsoft account credentials, according to Check Point researchers. The researchers identified 302,386 unique phishing emails, including bogus Teams invitations and .ics attachments that direct recipients to fake Microsoft sign-in pages. The attacks abuse familiar calendar features rather than exploit flaws in Teams, Outlook or the .ics format.
Attackers used more than 20,000 Cloudflare Pages domains over six months to display fake Microsoft security alerts and direct users to scam call centers. The campaign uses Help TDS to route traffic from compromised websites, malicious ads, social media and spam. During an Infoblox test, a call center operator instructed researchers to install UltraViewer, a legitimate remote access tool.
Microsoft patched CVE-2026-103572, a remote code execution flaw in Copilot affecting Visual Studio 2026, older versions including Visual Studio 2022, and SQL Server Management Studio. A trailing dot in filenames bypassed path validation, allowing Copilot to overwrite protected configuration files and run arbitrary commands. The researcher demonstrated exploitation through indirect prompt injection; Microsoft rated the flaw Important with a CVSS score of 8.8.
Hackers planted malware in some Rank Math PRO 3.0.123 WordPress plugin packages distributed October 15–16, compromising websites when activated and installing backdoors that can rebuild themselves after removal. Rank Math removed the affected packages and released clean version 3.0.124, but updating alone does not remove the infection. The free Rank Math SEO plugin is unaffected; the company did not disclose an affected website count.
Two websites, meet-google[.]cyou and qooglo[.]com, host fake Google Meet pages designed to steal Google credentials and facilitate two-factor authentication phishing. Researcher Anurag (@Malwarehunterr) also reported an exposed /install page revealing database configuration details and a prefilled database password. The report did not confirm any stolen accounts or database access.
F5 released emergency patches for three critical flaws affecting NGINX products, including NGINX Plus and NGINX Open Source, that can cause crashes, denial of service or code execution. CVE-2026-15176, CVE-2026-97205 and CVE-2026-93479 each carry a CVSS 4.0 score of 9.2 and depend on specific configurations. F5 said it found no evidence of exploitation in the wild.
Drupal patched CVE-2026-108097, a critical remote code execution flaw in its core Field UI module that allows unauthenticated attacks through certain contributed modules. Drupal core alone provides no direct exploitation path; custom code may also expose the flaw. Fixes are available in versions 10.6.19, 11.3.19, 11.4.10 and 12.0.0-beta2, and the advisory reports no attacks in the wild.
World has made World ID for Zoom generally available to enterprise customers, adding a “Human Badge” to verified participants’ video feeds to help distinguish people from AI deepfakes. Verification requires an iris scan at a physical Orb device and a selfie through World’s mobile app. Enterprises can choose to install an Orb onsite.
Varonis researchers gained operating-system command execution in Snowflake-managed PostgreSQL by chaining built-in features from a non-superuser account with CREATEROLE privileges. The chain used modified pg_cron jobs to gain database superuser access, then bypassed file-access restrictions to alter archive_command. Snowflake fixed the issue before publication; testing stayed within the researchers’ single-tenant environment, with no reported access to customer data.
An unpatched flaw in Broadcom’s Wi-Fi Protected Setup software lets attackers within radio range run code as root on routers without a password or pairing. Researcher Quentin Kaiser demonstrated exploits against gateways from Proximus, Orange Belgium and VOO and estimates millions of devices are affected worldwide. No CVE has been assigned, and Kaiser reports no real-world attacks.
North Korean IT workers used stolen or fake identities to secure remote jobs at more than 70 U.S. companies, generating about $3 million from 2019 to mid-2025, according to a federal indictment. U.S. authorities arrested two Americans and charged 14 people in coordinated enforcement actions. The FBI searched more than 40 Texas residences and seized approximately 187 computer devices used in the scheme.
Google released a Chrome update fixing 10 vulnerabilities across Windows, Mac and Linux, including a critical use-after-free flaw in Prerender, CVE-2026-110922. The fixes also cover CVE-2026-110913, a high-severity DOM memory corruption flaw reported by Anthropic’s Xinyang Ge with Claude AI assistance. Google is rolling out versions 155.0.8059.84/.85 for Windows and Mac and 155.0.8059.84 for Linux.
Attackers are increasingly impersonating AI brands such as Claude and ChatGPT in campaigns targeting enterprise employees, including developers. These campaigns typically use ClickFix-like tactics to target unsuspecting staff, according to the report.
Attackers compromised at least 143 devices in 34 countries, installing legitimate Pawns.app software without consent to monetize victims’ bandwidth through proxy networks. GreyNoise identified targets including D-Link ShareCenter NAS devices, Ubiquiti UniFi Video NVRs and SAP NetWeaver Visual Composer systems. Attackers exploited CVE-2024-3272 and CVE-2024-3273 in D-Link devices, CVE-2021-44228 in Ubiquiti systems and CVE-2025-31324 in SAP systems.