Skip to content

Image: research.checkpoint.com · rights & removal

Executive Summary

Recent cyber events involved data breaches at various entities, including TSMC and Maersk, where information concerning employees and system access was exposed due to unauthorized access. A data breach in Osaka, Japan, resulted from compromising a third-party service, exposing over one million email addresses. ConnectWise experienced attacks targeting its remote access platform, with attackers deploying malicious tools across multiple organizations. In the realm of AI threats, researchers observed an AI-assisted intrusion where scripts were generated by AI agents to extract credentials from monitoring software configurations. An autonomous campaign was noted where AI agents, specifically DeepSeek and Zhipu GLM, operated against numerous organizations, compromising sensitive information across diverse sectors. Vulnerabilities have been addressed through patches released by Microsoft for flaws like the Outlook flaw and code execution vulnerabilities, as well as Google patching a vulnerability in Chrome's XML processing component. Specific software vulnerabilities include an exploit chain involving PHP2Shell and CVE-2026-123039 against Check Point devices. Threat intelligence also points to cyberespionage linked to actors like CopperDragon and the use of AI agents by groups such as Shahid Kaveh.

Facts Only

* TSMC confirmed a data breach at an unnamed cloud provider exposing employee information.
* Maersk confirmed unauthorized access and data theft from a legacy system, involving the WhiteArmor extortion group claiming 100GB of stolen data.
* Osaka city government disclosed a data breach exposing over one million email addresses via a compromised third-party service.
* ConnectWise experienced attacks targeting its ScreenConnect platform across 48 organizations.
* Researchers observed an AI-assisted intrusion where attackers used AI-generated scripts to extract credentials from SolarWinds Orion configuration files.
* An autonomous intrusion campaign utilized DeepSeek and Zhipu GLM AI agents against at least 19 organizations over 25 operations.
* A server belonging to Silent Lynx was exposed, utilizing a DeepSeek-powered AI agent for reconnaissance.
* Check Point Research traced attacks exploiting PHP2Shell through Mobile Access portals of unpatched Security Gateways, using CVE-2026-123039 for privilege escalation.
* Microsoft released security updates fixing vulnerabilities including an actively exploited Outlook flaw (CVE-2026-124194) and code-execution flaws (CVE-2026-116928).
* Google patched CVE-2026-125350, affecting Chrome’s XML processing component.
* Apache fixed CVE-2026-119618 in HTTP Server versions 2.4.55 through 2.4.70 when HTTP/2 is enabled.
* Check Point Research identified cyberespionage campaigns linked to CopperDragon targeting government and critical infrastructure organizations.
* Researchers analyzed Russia-linked Turla’s Cereal toolset on compromised Exchange servers in Kazakhstan and Romania.

Full Take

The narrative presents a complex interplay between traditional large-scale data exfiltration, the emerging threat of AI-driven autonomous operations, and persistent, state-sponsored espionage. The incidents with TSMC and Maersk highlight that even established entities are vulnerable to legacy system compromises and extortion by organized groups. This juxtaposition is critical: sophisticated adversaries leverage both known vulnerabilities (like unpatched systems exploited via PHP2Shell) and novel capabilities (AI agents) simultaneously. The use of AI in reconnaissance, as seen with the Silent Lynx activity, suggests a shift from manual, targeted attacks to scalable, automated information gathering, which inherently lowers the barrier for large-scale espionage against geographically diverse targets.
The pattern emerges in how threat actors seek high-value data—employee records, credentials, and system configurations across critical infrastructure. The involvement of specific AI models (DeepSeek, Zhipu GLM) suggests a convergence where advanced reasoning is being weaponized not just for novel exploits but for efficient execution of known objectives like data exfiltration or evidence removal, blurring the line between sophisticated coding and autonomous decision-making. Furthermore, the context involving groups like CopperDragon and IRGC-affiliated units points toward geopolitical motivations driving the exploitation of these technical flaws to achieve strategic objectives, such as sabotaging physical infrastructure.
The implication for cognitive sovereignty lies in recognizing that defenses must account for both known patching cycles and the exponential growth of autonomous threat capabilities. If systems are being leveraged by AI agents capable of rapid adaptation during intrusion, reliance on static perimeter defenses becomes insufficient. The question is whether organizational structures can effectively govern or counter threats where the attack vector (AI) operates at a speed and complexity that outpaces traditional human response, and how this affects accountability when state-linked actors utilize these tools for espionage against disparate global interests. What mechanisms exist to establish resilience when the methods of attack are continuously evolving based on accessible large models?

From the original · Check Point Research

For the latest discoveries in cyber research for the week of 16th November, please download our Threat Intelligence Bulletin. TOP ATTACKS AND BREACHES - TSMC, a major Taiwanese semiconductor manufacturer, has confirmed a data breach at an unnamed cloud provider that exposed information about current and former employees.
Read the full story at research.checkpoint.com

Sentinel unavailable

The automated check of the source article's wording did not complete, so no result is shown.

This looks only at the wording of the original source article, not at this page's AI-written sections. A small local AI model made this estimate. It has not been checked against known human and machine texts, so treat it as provisional. It cannot show who wrote an article.

16th November | Huntaegis