Check Point: Attacks Up 48% YoY as AI Becomes Standard Offensive Tooling

Check Point Research published its weekly Threat Intelligence Report on October 12, 2026, logging a 48% year-over-year increase in global cyberattacks for September, averaging 2,803 weekly attacks per organization. The report, rich with verifiable intelligence across multiple fronts, draws a clear line: generative and agentic AI is no longer solely a defensive tool but a standard component of the offensive arsenal, with measurable effects on the scale, personalization, and surface credibility of attacks.

The Democratization of Offense: When AI Lowers Operational Cost

The October 12, 2026 Check Point Research report documents a structural shift in the threat landscape. The 48% year-over-year acceleration in global attacks for September — with peaks in the education sector, where reported ransomware attacks grew 53% — is not readable as mere cyclical intensification. The novelty lies in the convergence of open-source AI toolkits, commercial LLMs, and traditional command-and-control infrastructure.

The ARTEX campaign, documented by CrowdStrike Intelligence with direct technical evidence, is emblematic. Threat actors used open directories at IP address 38.244.50[.]120 to host toolkit instances, Claude Code session files, and a CLAUDE.md document containing Chinese-language pentesting prompts specifying how to conduct offensive activities. The presence of exposed session histories indicates an operational error — the failure to protect directories — that allowed researchers to reconstruct the tradecraft. CrowdStrike has not attributed the campaign to a named group, and the exact number of South Korean financial institutions hit remains unconfirmed.

The quantitative data on enterprise GenAI prompts — one in 39 with high risk of sensitive data exposure — measures an internal attack vector that traditional policies do not cover. This is not malware bypassing the perimeter, but legitimate interactions with enterprise tools that expose sensitive information through unintentional or malicious prompt engineering.

"The use of agentic AI tooling alongside traditional offensive capabilities highlights the continued evolution CrowdStrike has observed in adversarial tradecraft" — CrowdStrike Intelligence

CyberXero, AitM Phishing, and the New Social Engineering Chain

CyberXero, an initial access broker with AI-augmented capabilities, targeted Ukrainian energy infrastructure and WordPress/e-commerce platforms, obtaining data on more than 628,000 Ukrainian individuals. The combination of automated attack surface identification and personalized spear-phishing content reduces reconnaissance time and increases the success rate of social engineering campaigns.

A second campaign, analyzed by Cisco Talos, targeted Taiwanese research organizations with event invitations that used authentic details to mask malicious intent — a technique Talos defines as "legitimacy laundering." The phishing kit employs a hybrid HTTP/WebSocket architecture to synchronize authentication flows in real time, enabling interception of credentials and MFA codes. The interface was originally developed in Simplified Chinese and later adapted. Talos assesses the linguistic connection of the developer with "moderate confidence."

Simultaneously, Check Point Research documented hidden prompt injection campaigns in HTML comments and invisible text against employees and AI email assistants. This vector exploits the trust boundary between seemingly benign content and hidden instructions addressed to the LLM, with potential for data exfiltration or agent behavior manipulation.

High-Impact Breaches and Critical Vulnerabilities With Patches Available

The report logs multiple high-profile compromises in the reference period. The Danish government confirmed the breach of the Central Register of Persons with data on approximately 8.8 million individuals — names, addresses, and identification numbers — exploiting access granted to a private vendor. Ukrainian food retailer ATB, the country's largest, suffered an attack with an extortion notice on its site; the group DataSuckers claims theft of 7.9 million customer records and over 11 million order records, publishing samples on Telegram. ATB officially denied data compromise, stating that "the temporary message displayed on the website did not affect the security of your data." The authenticity of the DataSuckers samples has not been independently verified. The Record reports a ransom demand of approximately $400,000.

Osaka Metropolitan University suffered a suspected ransomware attack with approximately 500 servers hit and data of at least 130,000 individuals potentially exposed. Containment status at the time of the report is not specified.

The vulnerability front presents four critical patches. Veeam released updates for CVE-2025-64393, an RCE with CVSS 9.4 in Backup & Replication 12, allowing backup users with limited privileges to execute commands. Citrix confirmed targeted exploitation of CVE-2026-88779, a memory overflow in NetScaler ADC/Gateway with CVSS 8.7, tied to SAML authentication. Cisco patched CVE-2026-76455 in NX-OS, an improper access control with CVSS 9.8. Atlassian patched CVE-2026-21589, arbitrary file access with CVSS 9.3, affecting eight self-managed Data Center products.

The FortiBleed campaign, reported by the FBI and Secret Service, continues to cause damage with more than 86,000 FortiGate firewalls and VPN gateways compromised across 194 countries — a figure underscoring the persistence of unmitigated attack surfaces despite widespread disclosure.

Why It Matters

The Check Point Research dossier of October 12, 2026 does not document specific remediation measures for AI-assisted threats nor detailed operational recommendations for mitigating prompt injection risk. The source does not specify the full nature of data exposed in the Danish breach beyond names, addresses, and identification numbers, nor the exact perimeter of data obtained by CyberXero.

What the report documents with solidity is a phase transition: the barrier to entry for sophisticated attacks is eroding. Generative AI does not merely amplify scale; it amplifies surface credibility — the "presentation layer" of the attack — making criminal operations indistinguishable at first glance from campaigns attributable to nation-states. The gap between offensive adoption speed and defensive countermeasure maturity is measurable in the report's quantitative data and the technical evidence of documented campaigns.

Traditional defenses — signature-based detection, generic awareness, network perimeters — are not designed for autonomous agents that explore attack surfaces, generate personalized persuasive content, or inject prompts into legitimate channels. The transition demands a rethinking of enterprise GenAI policies that the report does not detail but whose need is implicit in the numbers presented.

Frequently Asked Questions

Is ARTEX attributable to a known group?
No. CrowdStrike Intelligence explicitly stated the campaign "has not been attributed to a named adversary." Attribution remains unconfirmed.

Is the data stolen from ATB authentic?
It has not been independently verified. ATB denies compromise; DataSuckers published samples on Telegram but the primary source does not confirm their authenticity.

Are the Taiwan phishing emails LLM-generated? Talos assesses "likely used AI-assisted" but states it cannot determine this conclusively. The primary Check Point Research report adds no confirmation on this specific aspect.

Information is based on the cited source and current as of publication.

Sources

• https://research.checkpoint.com/2026/12th-october-threat-intelligence-report/
• https://therecord.media/atb-ukraine-cyberattack-ransomware
• https://www.crowdstrike.com/en-us/blog/unknown-threat-actor-uses-artex-to-target-south-korean-finance/
• https://blog.talosintelligence.com/uat-11985/