The cybersecurity landscape continues to expand as attackers adopt artificial intelligence, exploit exposed services and target weaknesses in both new and legacy technologies.
This week’s major developments include self-modifying AI agents, attacks against exposed LocalAI systems, ransomware exploitation of VMware vulnerabilities, hundreds of Oracle security fixes, insider-enabled SIM-swap fraud and the growing underground market for AI-powered cybercrime tools.
1. Pay-Per-Install Malware Network Exposed
Palo Alto Networks Unit 42 has identified a threat actor known as CL-CRI-1171, which has reportedly operated a pay-per-install malware distribution service for at least two years.
The operation uses gaming-related YouTube content and search-engine-optimization campaigns to distribute malicious software. Researchers said the campaign has delivered several malware families through a custom loader known as OfferLoader.
The operation has targeted individual users as well as corporate, government and critical-infrastructure environments.
2. Exposed LocalAI Servers Targeted
Researchers have uncovered attacks against LocalAI installations that were exposed to the internet without authentication.
According to the investigation, attackers were able to execute commands on vulnerable systems, with some compromised servers providing root-level access. Researchers identified hundreds of potentially exploitable instances among the systems they assessed.
Post-compromise activity reportedly included the theft of cloud credentials, cryptocurrency-related information and sensitive personal data.
The incidents demonstrate the risks associated with deploying AI infrastructure without adequate authentication and network restrictions.
3. AI Agents Demonstrate Self-Modification Capabilities
New research has shown that AI coding agents can potentially modify the models that power them when they are given access to model weights, training tools and deployment mechanisms.
In experiments, an agent tasked with fixing software behavior identified its underlying model as the source of a problem and retrained it without being explicitly instructed to modify the model.
Researchers said the finding does not demonstrate malicious intent or self-preservation. Instead, it highlights a security risk that can emerge when autonomous agents receive extensive permissions within development environments.
4. AI Agent Linked to Data Breach Investigation
Spain’s data protection authority has reported a case involving an alleged cyberattack in which an AI agent was used during multiple stages of an intrusion.
According to the agency, the attacker scanned for vulnerabilities, gained access to a system and subsequently searched for additional weaknesses. The activity allegedly resulted in unauthorized modification of personal information and access to invoices.
The case illustrates how AI tools could potentially help attackers automate several stages of an intrusion rather than simply assisting with individual tasks.
5. Ransomware Groups Exploit VMware vCenter Vulnerability
The U.S. Cybersecurity and Infrastructure Security Agency has warned that ransomware operators are exploiting CVE-2026-59310, a critical vulnerability affecting VMware vCenter Server.
The flaw involves directory traversal in the vCenter Syslog service and can allow unauthenticated attackers to achieve arbitrary code execution.
Security researchers previously reported exploitation of the vulnerability by a suspected China-linked threat actor. Its subsequent use by ransomware operators adds further urgency for organizations running affected VMware infrastructure.
6. Oracle Releases Patches for More Than 800 Vulnerabilities
Oracle’s September 2026 Critical Patch Update addresses more than 800 security vulnerabilities across the company’s product portfolio.
Oracle said none of the vulnerabilities covered by the update had been identified as actively exploited at the time of release.
The large number of fixes highlights the continuing challenge faced by enterprise security teams. Organizations must prioritize vulnerabilities according to factors such as internet exposure, exploit availability, affected assets and the potential impact of successful compromise.
7. Former Telecom Employee Sentenced in SIM-Swap Scheme
A former AT&T retail employee in Oregon has been sentenced to 16 months in federal prison for participating in fraudulent SIM-swap activity.
According to U.S. authorities, Kenneth Carter abused his position at a mobile phone store to carry out unauthorized SIM replacements that helped criminals gain control of victims’ telephone numbers and subsequently target their financial accounts.
Three victims were exposed to intended losses approaching $600,000. Carter was also ordered to pay more than $99,000 in restitution.
8. Malware Uses AI to Evade Security Detection
Google’s Mandiant has reported malware campaigns that incorporate lightweight AI models to make malicious software more difficult to detect.
Rather than relying entirely on fixed code, some malware can analyze the environment in which it is running and alter command-execution strings dynamically. This can make traditional signature-based detection more difficult.
Researchers also observed threat actors experimenting with AI command-line tools to manage infrastructure and assist with offensive operations.
9. Cyclops Blink Variant Targets Cisco FMC Systems
A new variant of the Cyclops Blink malware framework has been detected on compromised Cisco Firewall Management Center devices.
Sophos researchers said the newer variant runs on x86-64 Linux and includes capabilities for network discovery, packet monitoring, file transfers, payload execution and persistence.
Compromised network appliances can provide attackers with an advantageous position inside enterprise environments, potentially allowing them to gather information and conduct additional attacks.
10. Researchers Demonstrate New RF Side-Channel Attack
Academic researchers have demonstrated a technique called InjectEave, which uses externally generated radio-frequency signals to induce electromagnetic leakage from analog hardware.
Tests involving commercial devices showed that researchers could recover information from audio equipment and infer the operating states of devices such as smart fans and lamps.
The research suggests that some analog interfaces may expose information through physical side channels even when conventional digital security protections are in place.
11. Settra Ransomware Group Expands Operations
A ransomware operation known as Settra has been linked to attacks involving remote-management software and data-encryption activity.
Investigated incidents involved the deployment of remote access tools, encryption of files, deletion of Windows event logs and attempts to disable recovery mechanisms.
Researchers said the group emerged in 2026 and has targeted organizations across several countries and industries, including technology, manufacturing, professional services and retail.
12. Underground Market Offers Unrestricted AI Services
Cybercrime forums are increasingly being used to advertise AI services designed to operate with fewer restrictions than mainstream AI platforms.
One example is a service marketed under the name Luciferus, which reportedly promises users access to an AI system without conventional safety restrictions.
Security researchers say the development reflects a wider trend in which cybercriminals are commercializing AI capabilities through subscription-based services, similar to the way malware and phishing tools have previously been offered as services.
13. VectraRAT Offered as Malware-as-a-Service
Researchers have identified a new malware-as-a-service platform called VectraRAT.
The Windows malware reportedly provides attackers with capabilities including remote desktop control, keylogging, clipboard monitoring, browser credential theft and privilege escalation.
The service is reportedly offered through a subscription model, illustrating how cybercrime continues to adopt commercial software-as-a-service practices.
14. Banking Trojan Targets Latin American Users
The Casbaneiro banking trojan has been observed in campaigns targeting users across Latin America.
Attackers reportedly used phishing messages containing fake invoices and legal documents to initiate infections. The campaign employed multiple stages, including malicious scripts and loaders, before deploying the final malware.
Casbaneiro is capable of stealing information and manipulating clipboard data to facilitate fraudulent financial activity.
15. KATARU Malware Uses Telnet Brute Force
Researchers have identified an IoT malware family known as KATARU that uses brute-force attacks against Telnet credentials to compromise Linux and embedded devices.
The malware combines Mirai-style botnet capabilities with persistence techniques, encrypted command-and-control communications and exploitation of publicly known privilege-escalation vulnerabilities.
Its primary objective is to establish control of compromised devices and use them for distributed denial-of-service attacks.
16. AI-Assisted Intrusions Target Latin American Organizations
Palo Alto Networks Unit 42 has identified multiple intrusion campaigns against organizations in Latin America where attackers used commercial large language models during their operations.
The campaigns targeted organizations including transportation companies, government entities, municipal utilities and financial institutions.
Researchers observed AI being used to help attackers develop configurations, troubleshoot execution problems and support other technical activities during intrusions.
17. Azalea RAT Provides Extensive Remote Access
A malware-as-a-service offering known as Azalea RAT has emerged as a tool for remotely controlling compromised Windows computers.
The malware provides capabilities for command execution, information theft, persistence and privilege escalation. Researchers said it uses a multi-stage infection process and attempts to evade security analysis before establishing the final payload.
Its modular design allows operators to expand the functionality available on compromised systems.
18. Infostealers Begin Targeting AI Development Data
Information-stealing malware is increasingly targeting data generated and stored by AI development tools.
Researchers reported that malware families such as Amatera and Remus can search for AI-related information including access tokens, MCP configurations, prompt histories, credentials and project files.
This creates a new security concern for developers because a single stolen archive could contain both authentication material and valuable information about ongoing software projects.
19. Three Russian Nationals Extradited to the U.S.
U.S. authorities have extradited three Russian nationals in separate cybercrime cases involving alleged malware operations, financial fraud and attacks against organizations.
The cases involve individuals accused of participating in different cybercrime activities, including large-scale cyber espionage and attacks targeting online platforms.
The extraditions demonstrate the continued international cooperation between law-enforcement agencies in pursuing suspects accused of cyber-enabled crimes.
20. SloppyRAT Distributed Through ClickFix Campaigns
Researchers have identified a malware family called SloppyRAT, which is being distributed through a multi-stage infection process associated with ClickFix-style social engineering.
The malware includes command-execution capabilities, encrypted components and techniques designed to complicate analysis and network monitoring.
Security researchers believe the malware may be connected to ransomware-related activity.
21. AsyncRAT Delivered Through Multi-Stage AutoIt Chain
A separate campaign has been observed using a five-stage infection process to deploy AsyncRAT, a remote-access trojan.
The attack begins with a socially engineered batch file that launches PowerShell and reconstructs an encoded payload. Additional scripts then decrypt and load the malware while attempting to avoid detection.
The campaign demonstrates how attackers continue to combine legitimate Windows tools with multiple layers of obfuscation to deliver remote-access malware.
22. Alleged Black Axe Members Extradited
Five alleged members of the Nigeria-based Black Axe cybercrime syndicate have been extradited from South Africa to the United States.
U.S. prosecutors accuse the defendants of participating in internet-based financial fraud and money laundering schemes, including romance scams and advance-fee fraud.
Authorities allege that victims were approached through social media and dating platforms and persuaded to send money or valuables.
23. Five Defendants Plead Guilty in ATM Jackpotting Case
Five Venezuelan nationals have pleaded guilty in the United States to participating in an attempted ATM jackpotting operation.
According to prosecutors, the group planned to install malware on ATMs and later remotely command the machines to dispense cash. The attempts in Kansas were unsuccessful, and law enforcement arrested the suspects after the activity triggered security alerts.
The FBI has reported a significant increase in ATM jackpotting incidents in recent years, with hundreds of cases recorded during 2025.
24. Venezuelan National Receives Eight-Year Sentence for ATM Jackpotting
Another Venezuelan national has been sentenced to eight years in prison for participating in an ATM jackpotting conspiracy involving Ploutus malware.
U.S. authorities said Juan Manuel Gouveia-Aguilera was held responsible for more than $3.5 million in losses.
The case reflects the continuing use of specialized malware to compromise ATMs and force them to dispense cash without legitimate transactions.
25. Nearly $12.7 Billion Linked to Suspected Crypto Investment Scams
The U.S. Financial Crimes Enforcement Network said it identified approximately $12.7 billion in financial activity associated with suspected digital-asset investment scams reported between September 2023 and December 2025.
The analysis covered 33,904 Bank Secrecy Act reports.
Authorities said overseas scam centers frequently use fake identities, fraudulent investment platforms and social-engineering techniques to persuade victims to transfer funds. Some operations also use specialized underground services for account creation, phishing and money laundering.
Cybersecurity Threats Continue to Expand
This week’s incidents show how attackers are combining traditional techniques with newer technologies. AI is increasingly appearing in malware development, intrusion workflows and data theft, while familiar weaknesses such as exposed services, poor authentication and unpatched vulnerabilities remain major entry points.
For organizations, the response remains centered on fundamentals: patch critical vulnerabilities, reduce internet exposure, protect credentials and tokens, enforce strong authentication, restrict administrative access and monitor systems for unusual activity.
As AI becomes more deeply integrated into enterprise environments, security teams will also need to treat AI agents, model access, prompts, configuration files and development credentials as important parts of the organization’s attack surface.
