Artificial Intelligence Embeds Deeply Into Global Cyber Threat Landscape as New Attack Campaigns Target Organizations Across Japan

Cybersecurity researchers have disclosed detailed findings regarding three distinct cyber attack campaigns targeting organizations in Japan, where artificial intelligence served as a foundational element. In these operations, AI was deployed either to take the driving seat in automating the entire attack lifecycle or to act as an advanced assistant to human operators, marking a significant evolution in how threat actors weaponize modern technology.

According to a comprehensive analysis by TrendAI, which inspected three exposed server environments utilized by threat actors to conduct their malicious operations, these incidents differ fundamentally from previously documented attacks. Earlier reports have highlighted the use of ARTEX, an autonomous AI penetration testing framework, to single out specific corporate and institutional entities in South Korea and Japan. However, the newly uncovered environments displayed distinct characteristics, holding a rich repository of reconnaissance scripts, target lists, operational results, logs, and command histories.

Security researchers Bakuei Matsukawa and Joseph C. Chen noted that these exposed workspaces varied significantly in their operational setup. They ranged from environments manually controlled by skilled human operators utilizing large language models for acceleration to fully automated AI-agent platforms capable of executing the entire workflow, including command and control, with minimal human intervention.

The Rise of Automated CyberStrikeAI Operations

The first of the three newly detailed attacks involved the abuse of CyberStrikeAI, an open-source penetration testing tool that had previously made headlines for being deployed in targeted campaigns against Fortinet FortiGate appliances. This specific wave of activity was concentrated within a remarkably short period during October 2026.

The AI-driven operational framework leverages this offensive security tool to automate various facets of the traditional attack lifecycle into a seamless, highly repeatable workflow. The automated sequence begins with broad reconnaissance and vulnerability discovery, followed by validation, exploitation, command-and-control establishment, persistence, data exfiltration, and the automated generation of detailed post-operation reports.

This level of automation involves delegating complex tasks among 44 distinct AI agents to handle specialized aspects of the intrusion. Furthermore, the platform empowers threat actors to dynamically generate web forms and convincing fake login pages tailored specifically in the Japanese language, increasing the efficacy of credential harvesting efforts.

Trend Micro’s enterprise security division highlighted the systematic nature of these intrusions, noting that within a very brief timeframe, at least 19 identifiable organizations—including a minimum of 16 located in Japan—were subjected to systematic validation activity. Each targeted entity had a corresponding report generated in an identical format, complete with an index of collected evidence. Investigators emphasized that these structured outputs were the direct product of a rigorous platform-run process rather than ad-hoc notes left behind by human operators.

Human-Led Intrusions and LLM-Assisted Infrastructure Targeting

In contrast to fully automated agent platforms, the second case identified by researchers has been characterized as a human-led attack that utilized large language models merely to accelerate strategic planning and tool development. An exhaustive analysis of the infrastructure and digital artifacts associated with this campaign revealed that its objectives extended beyond traditional data theft. The operation was specifically designed to target AI workflow platforms and inference services, with the ultimate goal of hijacking and acquiring large amounts of computational inference capacity at a remarkably low cost.

One notable intrusion within this campaign targeted a geographic information system server that had been left exposed with default administrative credentials. This oversight resulted in the rapid theft of sensitive cadastral and commercial records, Windows user credentials, and internal email data. The entire campaign maintained persistence and executed its objectives over a 24-day period spanning August and September 2026.

The third instance featured a more conservative application of artificial intelligence, where the technology’s role was strictly confined to developing the operator’s custom tools and scripts. This limited integration supported a series of persistent campaigns designed to target edge networking and gateway devices manufactured by Cisco, Citrix, F5, and Yamaha. The threat actors exploited published software vulnerabilities, management functions, or administrative credentials inadvertently left present in plain-text configuration files.

Investigators estimate that the operator behind these edge-device intrusions ran at least six distinct campaigns between January and October 2026, opportunistically striking exposed virtual private network devices and WebLogic servers. Researchers described the operator as highly experienced, possessing established practices for operational automation and evidence management.

Global Reach and Evolving Threat Dynamics

Currently, cybersecurity investigators have found no conclusive evidence to suggest that these three distinct incidents share a common operator. Furthermore, it remains unknown whether these specific threat actors expanded their targeting to other global regions or precisely which underlying commercial or open-source AI models were utilized to orchestrate the attacks.

Nevertheless, the geographic footprint of these campaigns extends well beyond Japan and neighboring South Korea. The threat actor responsible for the first case was observed targeting organizations in Taiwan and Hong Kong, while the third case incorporated organizations located in the United Kingdom, France, Italy, Qatar, and Colombia within its WebLogic target discovery lists.

Commenting on the broader strategic shift, TrendAI observed that artificial intelligence has effectively expanded the workforce available to threat actors. However, researchers noted that the fundamental nature of the intrusions remains unchanged—attackers are still primarily searching for unlocked doors and known weaknesses. What has fundamentally changed is the capability to test thousands of potential entry points simultaneously, adapt malicious content to target country languages, and sustain operations without human fatigue.

This development aligns with broader intelligence findings across the cybersecurity industry. Multiple recent campaigns have demonstrated that threat actors of diverse motivations are successfully integrating agentic AI into existing offensive security workflows. This integration allows malicious actors to dramatically accelerate research, malware development, social engineering, and other critical phases of an attack lifecycle.

Security analysts warn that AI is accelerating reconnaissance, automating complex decision-making, and industrializing social engineering techniques, thereby drastically shortening the window of vulnerability between the public disclosure of a security flaw and its active exploitation. According to Microsoft’s Digital Defense Report, the median time required for threat actors to transition from vulnerability discovery to operational weaponization has collapsed to well below 24 hours.

Industry leaders note that the widespread adoption of artificial intelligence by malicious actors significantly reduces the time, technical expertise, and financial cost required to discover and exploit network weaknesses. At the same time, it enables both attackers and defenders to operate with unprecedented speed, scale, and operational autonomy.

Security experts emphasize that the core security challenge is no longer merely whether a threat actor possesses access to frontier AI models. Access to moderately capable models is already sufficient to yield meaningful operational gains. Instead, ecosystem depth—how reliably, independently, and at what scale those capabilities can be sustained—now dictates the effectiveness of modern cyber operations. Artificial intelligence has transitioned from an experimental capability for advanced persistent threat groups into an operational component embedded directly into daily cyber campaigns.

The escalating demand for specialized AI tools among cybercriminals has simultaneously spawned a lucrative new monetization avenue within underground digital marketplaces. Cybercrime forums increasingly peddle stolen AI provider session tokens alongside specialized offensive AI frameworks such as APEX AI, MessiahGPT, and DarkGPT. This trend further commoditizes advanced cybercrime, significantly lowering the technical barrier of entry for malicious actors seeking to execute sophisticated, persistent, and evasive campaigns.

Recent enforcement and intelligence disclosures underscore the urgency of these developments. Authorities in South Korea recently identified seven suspects linked to an extensive cybercrime operation that breached more than 90 organizations, largely relying on the ARTEX framework. Investigations revealed that the 32-year-old mastermind behind the syndicate previously worked at a prominent cybersecurity firm, leveraging professional penetration testing expertise for illicit gains. In dozens of those incidents, the threat actors demanded substantial ransom payments in exchange for withholding stolen data, highlighting the severe commercial impact of AI-accelerated cyber attacks.