Attack Methods & Trends

Autumn 2026 brought Japan its largest recorded wave of data breaches and unauthorized access. NPA statistics show 123 confirmed ransomware cases in H1 2026 — a half-year record since counting began in 2020. Suspicious access detections reached about 13,700 per day (up ~50% year on year), and phishing reports topped 730,000.

The government declared cyberspace a state of emergency and ordered ministries, municipalities, and businesses to enforce basics (updates, strong passwords, supply chain checks). Authority to neutralize attacker tools under the Active Cyberdefense Law took effect October 1, 2026. Businesses can now consult the IPA's special unauthorized-access desk from the moment they suspect damage, and the PPC has issued alerts.

By technique, supply-chain cascades through shared services and vendors stand out domestically. Compromises of reservation systems, e-commerce builders, and clouds have hit hundreds of client companies and municipalities at once.

Vulnerability exploitation is now the top initial access vector. Verizon's 2026 DBIR puts it at 31%, overtaking stolen credentials. Exploitation of VPNs and edge devices jumped from ~3% to 22% — slow patching of perimeter devices is a worldwide weakness. Phishing is flat at 16%, pretexting grew to 6% as a path into ransom and extortion, and ransomware is involved in 48% of all breaches.

AI-enabled attackers are surging. CrowdStrike's 2026 Global Threat Report ("The Evasive Adversary Wields AI") counts an 89% year-over-year increase. Average breakout time hit a record-fast 29 minutes, and pre-disclosure zero-day exploitation rose 42%. Ransomware variants sharing AI-generated code defects have been observed — defenders are being tested on containment speed.

AI agents now steal directly. In South Korea, AI tools are suspected in attacks on several banks, and security researchers reported an AI agent stealing about 600,000 credit card records. Cheaper attacks mean mid-sized and small businesses everywhere are becoming targets. The Chinese AI systems reported in the 2026 attacks including Japan's are covered in the next section.

ARTEX (Chinese AI agent). In October 2026, an investigation by a Japanese company (a GMO subsidiary) confirmed that a Chinese AI tool, ARTEX, was used on all six servers behind a series of unauthorized accesses spanning roughly 80 Japanese companies in insurance, finance, and logistics. ARTEX originated as an open-source AI penetration-testing tool; industry-targeted operations against finance, retail, insurance, and e-commerce have been reported, and its abuse is confirmed or suspected in attacks on major Korean financial institutions and US companies. The developer withdrew it after the reports, but copies have spread and the threat persists. Multiple outlets report that a substantial share of Japan's autumn 2026 breach wave may be of this lineage — timing that matches the ~2.5× month-over-month surge in incidents we collected for October.

DeepSeek (Chinese open-source LLM). In July 2026, Palo Alto Networks' Unit 42 reported a Mandarin-speaking actor using DeepSeek via the "Hermes Agent" framework to run autonomous attacks combining AI-driven scanning of seven vulnerability classes with manual exploitation. Bloomberg (August 2026) reported that Chinese hackers have intensified activity since folding DeepSeek and other open-source models into their workflows. CSIS had already flagged DeepSeek's weak guardrails (jailbreak resistance, malware generation) in 2025: the pattern of cheap, loosely restricted Chinese models lowering the technical bar for attacks has persisted for over a year.

What the AI actually does. In every case, the AI handles preparation: scanning public services for vulnerabilities, generating and tuning attack code, and selecting targets — the final intrusion and theft still require human hands. The essentials of defense are unchanged: assume-compromise design (MFA, least privilege, log preservation, rapid detection). The AI era does not demote the basics.