The BOSSAI campaign involves a single operator using a mesh of 12 AI agents to conduct persistent attacks across 11 countries. The attacker leveraged both abliterated and commercial AI models in specialized roles such as infrastructure building, vulnerability scanning, and exploitation. Primary targets included emergency services, universities, financial institutions, and government systems in Taiwan, Thailand, and Mexico. Confirmed breaches include exfiltration of medical records from Taipei emergency services, student data from Thai universities, and over 200,000 Mexican residential phone records. Despite attempts to jailbreak commercial AI models, the attacker relied on less effective abliterated models. The investigation uncovered the full operational infrastructure, including command histories and victim data, highlighting both the capabilities and limitations of AI-augmented autonomous exploitation.
A Korean-speaking threat actor linked with moderate confidence to North Korea's Lazarus Group is conducting a global campaign exploiting two critical PHP vulnerabilities (CVE-2026-109618 and CVE-2026-109619) that allow unauthenticated remote code execution. The campaign targets government agencies, financial institutions, manufacturing, logistics, aviation, and trading platforms across multiple countries including the US, Japan, and EU. Exploitation attempts have been observed since 17 October 2026, involving web shell deployment and session poisoning. The vulnerabilities affect all internet-facing PHP installations below versions 8.5.12, 8.4.27, 8.3.36, or 8.2.35, with Apache mod_php being at highest risk.
Between April and July 2026, AOMEI Backupper, a Windows backup and disk cloning software, was compromised multiple times in short-lived supply chain attack windows. During these brief periods, users received a downloader for the ShinyRoad modular backdoor framework. The attacker, tracked as UNC7017, tampered with update server configurations to redirect users to malicious payloads, including the CrossRoad dropper, a modified Radmin Server with a malicious DLL, and the Adaptix C2 framework. The campaign used ephemeral tampering windows lasting only five to ten minutes, reverting quickly to authentic versions, which complicated detection. The attack affected a broad range of users opportunistically, with some higher education, IT services, and telecommunications organizations in multiple countries identified as compromised. AOMEI initially dismissed reports of tampering but later took corrective actions after repeated notifications.
Between April and July 2026, attackers compromised AOMEI's update servers to distribute malware via brief modifications to configuration files used by AOMEI Backupper's built-in promotional and notification features. The attackers altered two configuration files to redirect software updates to malicious payloads, delivering a downloader named CROSSROAD and a backdoor called SHINYROAD. These changes were short-lived, lasting only minutes, making detection difficult. The malware established persistence and enabled remote control of infected systems. Additionally, a technique to steal Windows authentication hashes was observed during one attack window. Affected organizations spanned multiple countries and sectors. AOMEI has since upgraded its servers, patched vulnerabilities, and taken corrective actions, though details on the initial breach vector remain undisclosed.
The Hotejka affiliate program, a rebrand of ROI777, operates the Help Traffic Distribution System (TDS) to deliver tech support scams. It uses browser fingerprinting and geolocation to redirect victims through compromised websites and malvertising to fake Microsoft security alerts that lock browsers and display fraudulent phone numbers. Victims who call are guided to install remote desktop tools like UltraViewer, granting attackers ongoing system access. Since April 2026, Hotejka has abused Cloudflare Pages to create over 20,000 disposable domains, generating about 300 new domains daily. Affiliates such as Balada Injector, Detour Dog, and DollyWay drive traffic to this TDS. Infoblox developed predictive detection algorithms that identify malicious domains about seven days before deployment, significantly reducing exposure.
A coordinated disruption targeted Chinese-speaking cybercriminals operating three major phishing kits—CoGUI, CoAceV, and CoFish_Ws—that distributed hundreds of millions of credential phishing messages primarily against Japanese organizations. These phishing campaigns impersonated 66 well-known consumer and finance brands to steal usernames, passwords, and payment data. The threat actors used sophisticated evasion techniques, including approximately 17,000 unique rotating domains. The disruption involved civil lawsuits and domain seizures against ecosystem enablers YAM Store and Iron Man, conducted collaboratively by Microsoft, Health-ISAC, and security partners. This action is expected to significantly disrupt multiple threat groups involved in financial fraud within the Chinese phishing ecosystem.
Microsoft uncovered a Chinese-language cybercriminal ecosystem involving phishing kit developers, an illicit marketplace called YAM Store, and downstream threat actors. This ecosystem enables large-scale phishing campaigns targeting organizations worldwide, including in Japan, the US, Australia, and India. The phishing kits impersonate legitimate brands and facilitate credential theft. YAM Store sells fraudulently acquired Microsoft and Google cloud environments to support these operations. Microsoft and partners disrupted the primary developers and infrastructure behind this ecosystem.
EvilTokens was an AI-powered phishing platform that enabled cybercriminals to compromise over 12,000 email accounts across more than 10,000 organizations worldwide by abusing device code authentication flows. The platform used AI to craft targeted phishing messages, select victims, and extract valuable data from compromised accounts. Microsoft disrupted the platform by seizing infrastructure and arresting two suspects linked to its operation. The platform charged users for access and offered numerous phishing themes to increase success rates.
A suspected Chinese-speaking threat actor linked to Red Heron conducted extensive exploitation campaigns from May to September 2026. The attacker exploited multiple vulnerabilities in WordPress (CVE-2026-63030, CVE-2026-60137) to breach a western government organization, stealing over 18,000 sensitive records including credentials and personally identifiable information. Additionally, the actor exploited CVE-2026-7273 in ZyXEL GS1900 switches, compromising 996 devices across 48 countries and exfiltrating configurations and credentials. The adversary demonstrated sophisticated capabilities such as custom tool development possibly using large language models, privilege escalation, and systematic data exfiltration via PowerShell-based techniques.
An Israeli influence-for-hire company called BlackCore has been identified conducting digital manipulation campaigns across multiple countries. The company operates through a sophisticated infrastructure offering services including discourse dominance, organic engagement manipulation, and counter-operations. Researchers identified a specific campaign involving a 14-week training program delivered to Angolan government employees in early 2026, which included the creation and deployment of fake social media personas and coordinated inauthentic behavior. The operation utilized AI-generated profile pictures, fake news outlets like 'Agita News', and coordinated amplification tactics across Facebook, Instagram, and TikTok. BlackCore's promotional materials openly advertised their capabilities to conduct deceptive influence operations on behalf of government clients, demonstrating how influence-for-hire services have become accessible to state actors seeking to manipulate online discourse.
MacSync Stealer is a macOS information stealer and remote-access stager delivered through ClickFix social engineering and malvertising campaigns. Operating under a malware-as-a-service model, it employs sophisticated evasion techniques including process daemonization, single-byte XOR obfuscation, and in-memory AppleScript execution to bypass Apple Gatekeeper, XProtect, and EDR solutions. The malware exfiltrates credentials, browser data, cryptocurrency wallets, and SSH keys through fault-tolerant 10MB chunked uploads to command-and-control infrastructure. MacSync targets professionals in software engineering, cryptocurrency, fintech, and corporate environments across North America, Europe, and Asia-Pacific regions, with campaigns impersonating legitimate services like Google Meet, Claude AI, Docker, and TradingView to trick victims into executing malicious Terminal commands.
Mathspace, an online mathematics platform, suffered a data breach impacting over 1 million individuals due to exploitation of a critical SQL injection vulnerability (CVE-2026-72898) in its self-hosted Metabase instance. The breach involved unauthorized access and data exfiltration of personal information including names, user IDs, usernames, email addresses, and related metadata of students, teachers, staff, and parents/guardians primarily from Australia and New Zealand. No sensitive academic records, passwords, or authentication credentials were exposed. The incident was discovered weeks after exploitation began, and remediation was delayed beyond the patch release date. Mathspace has taken steps to secure the environment and is investigating the incident while notifying affected individuals.
Three distinct implants—SPEAKINGSTONE, DARKLANTERN, and ENDLESSDOORS—have been discovered embedded in ZBT router firmware distributed through a global supply chain reaching the United States, Canada, Australia, Philippines, Germany, and Russia. DARKLANTERN is an unauthenticated UDP backdoor listening on port 9992, offering root shell access from the internet with trivial MAC address bypass. SPEAKINGSTONE is a phone-home surveillance implant that beacons to ZBT's cloud infrastructure, capable of DNS hijacking, ISP credential theft, and remote command execution. A sinkholed backup domain revealed 392 devices, 390 located in China, primarily on China Mobile's network. Internet scans identified 203 DARKLANTERN instances across 22 countries. These implants use plaintext protocols without authentication, making them hijackable by any network adversary. The affected hardware appears in white-labeled consumer products sold through mainstream retailers, representing embedded surveillance capabilities in commerciall...
The Iranian-linked cyber-espionage group Tortoiseshell has expanded its malware toolkit with a new reverse SSH tunneling utility disguised as wtsapi32.dll and a C++ backdoor similar to TWOSTROKE malware. The SSH tunnel uses the Windows OpenSSH client to connect to command-and-control servers. The backdoor supports executing files, shell commands, in-memory DLL execution, and file manipulation. Infrastructure analysis shows domains linked to multiple countries including the UAE, Saudi Arabia, UK, Belgium, Canada, Australia, Japan, and the United States, indicating broader targeting beyond the Middle East and US defense and military sectors. The group has been active since 2018 and maintains persistent infrastructure despite domain suspensions. No known exploits in the wild or patches are reported.
A sophisticated traffic distribution system has been operating for over 14 months, using more than 12,700 structurally similar fake CAPTCHA PDFs hosted on Webflow's CDN. The operation begins with search engine optimization, where victims searching for legitimate content encounter malicious PDFs through Google searches. These documents contain fake CAPTCHA panels that route users through a custom Elixir/Phoenix traffic distribution system employing IP filtering, bot detection, and geographic targeting. The infrastructure sorts visitors and redirects qualifying traffic to three distinct endpoints: Legion Loader distribution, a TDS reseller gate, and premium-SMS subscription scams targeting Spanish-speaking users. Non-qualifying traffic is monetized through search-arbitrage advertising. The operation primarily targets English-speaking countries and has recently been surfaced by AI assistants including Google Gemini and Claude, expanding its reach beyond traditional search engines.
The Greatness phishing-as-a-service (PhaaS) platform targets Microsoft 365 accounts using advanced phishing techniques including adversary-in-the-middle and device-code phishing. It spoofs RingCentral emails to bypass email security filters by exploiting whitelisting and safe-sender list trust. Post-compromise, attackers access Microsoft 365 data and services via stolen authentication tokens. The platform is sold to cybercriminals and has been active since at least mid-2022, targeting users in multiple countries. Researchers recommend auditing safe-sender lists and monitoring suspicious MFA-approved sign-ins.
The OpenAI models targeted services beyond Hugging Face as they attempted to solve the tasks they were given. The post JFrog Zero-Days Exploited in OpenAI-Hugging Face Hack appeared first on SecurityWeek .
In June 2026, infrastructure pivoting from TencShell C2 nodes revealed an active intrusion campaign utilizing AI language models for attack automation. Thirteen Hong Kong-based servers across four ASNs exposed an open directory containing victim source code, custom exploits, operational logs, and cloned login pages with notes in Simplified Chinese. The operation employed Claude Code for execution and DeepSeek-v4-pro for attack logic, targeting government systems in Afghanistan, Thailand, and Taiwan, along with reconnaissance against U.S. government portals. The campaign also pursued financial services firms across Europe, Australia, and Asia. Attackers deployed TencShell implants, webshells, and custom exploits including SQL injection and Laravel deserialization attacks, successfully compromising administrative systems and exfiltrating sensitive data including citizen complaints and government employee information.
A misconfigured Python HTTP server on a Budapest VPS exposed the complete operational infrastructure of three distinct phishing operators. The investigation uncovered codemado, an Egyptian threat actor operating since 2018, running a full AiTM platform with custom tools including MaDoO Blaster; saroula01, deploying OAuth Device Code Flow attacks that accumulated 218 victims across 12 countries over a year; and mail-argenta, a Nigerian operator identified through infostealer logs containing his own credentials. All three actors leveraged customized Evilginx forks and AI-assisted development to build MFA-bypass infrastructure from public GitHub repositories. The campaigns targeted Microsoft 365 accounts primarily, with codemado maintaining ties to RockyBelling's "The Quarry" cybercrime ecosystem. The exposed server contained phishing configurations, credential logs, RMM installers, combolists, and Telegram session files, revealing sustained operations from at least January 2025 through May 2026.
The Australian Cyber Security Centre (ACSC) has issued an alert about a global exploitation campaign targeting multiple vulnerable content management systems (CMS) and their plugins. The campaign involves deploying webshells on compromised websites, enabling persistent unauthorized access. Several WordPress plugins and CMS platforms including Craft CMS, MaxSite CMS, MetInfo CMS, and Joomla JCE are affected. The campaign has impacted many small- to medium-sized Australian businesses. The ACSC warns that AI may be used to accelerate and scale these attacks. Recommended mitigations include applying the latest security updates, removing unused components, enabling automatic updates, and restricting access to sensitive directories.
