Open-Source AI Pentesting Tool Weaponized to Hack 80 Japanese Companies
A Chinese engineer open-sourced an AI agent to find network bugs. Hackers weaponized it to breach Asian financial systems and corporations.
Tools · Source: Wall Street Journal
What happened
Hackers are deploying an artificial intelligence agent named ARTEX to breach corporate networks across Asia. Tokyo-based GMO Cybersecurity discovered digital traces of the tool in recent attacks against roughly 80 Japanese companies. The targets include major players like Daiwa Securities, SoftBank, and East Japan Railway. The attacks were highly organized. Attackers set up six distinct AI-driven server infrastructures, each focused on specific sectors like finance, logistics, and e-commerce. Eight of the targeted companies have already disclosed severe data leaks in the past two months.
ARTEX was never intended to be a weapon. Chinese cybersecurity engineer Li Puhua developed the system as an autonomous penetration testing tool. He released it for free on GitHub in July. His goal was to help defenders identify network vulnerabilities before malicious actors could exploit them. Instead, hackers immediately adapted the code. They used ARTEX to manipulate powerful large language models from Anthropic and DeepSeek. The agent now hunts for internet bugs around the clock without requiring human intervention or sleep.
The fallout is spreading rapidly beyond Japan. Before these recent incidents, ARTEX was detected in cyberattacks against seven South Korean financial firms. This marked one of the first known instances of an AI agent invading the global financial system. Microsoft also reported that a China-aligned hacking group used the tool to target hundreds of organizations worldwide. The original developer pulled the code offline in October, stating its malicious use violated his intentions. However, pirated versions continue to circulate widely on the dark web.
Key facts
- 80 — Japanese companies targeted by attacks bearing ARTEX traces
- 7 — South Korean financial firms attacked using the same AI tool
- 8 — Targeted Japanese companies that disclosed data leaks in the past two months
- 6 — AI-driven attack servers found targeting specific Japanese industries
Why it matters
The barrier to entry for sophisticated cyberattacks just dropped to zero. AI agents fundamentally change the economics of hacking. An AI tool like ARTEX can take a single published vulnerability, trace it through every version of an application code base, and find unflagged copies of the exact same flaw. Hackers no longer need massive teams or deep technical expertise to execute complex breaches. They only need an open-source script and API access to a frontier model. This makes cyberattacks infinitely cheaper and more efficient than conventional methods.
The line between defensive and offensive AI tools is officially dead. If you build an autonomous agent to patch systems, you have simultaneously built an agent to break them. Companies can no longer rely on static defenses when attackers deploy tireless AI agents that probe infrastructure twenty-four hours a day. The International Monetary Fund recently warned that these advanced AI programs represent a severe new threat to global financial stability. Infrastructure that is not prepared for autonomous attacks will be dismantled.
For builders
Assume open-source security tools will be weaponized
If you build AI agents for security or penetration testing, bad actors will fork them immediately. You cannot control how open-source code is deployed in the wild. Builders must implement hardcoded safeguards or risk becoming the engine for a global cyberattack.
Prepare for autonomous AI penetration attacks
Traditional rate limits and firewalls will fail against AI agents that mimic human traffic while hunting for bugs. Infrastructure teams must upgrade to AI-driven defense systems immediately. Companies relying on legacy security protocols will pay the ultimate price in data breaches and lost customer trust.
Secure your model API access points
Attackers are using legitimate models from Anthropic and DeepSeek to power their malicious agents. AI companies must improve their monitoring to detect when their models are being manipulated for automated network exploitation. If they fail, regulators will step in and force compliance.
My take
We built AI to protect us, and now the attackers are using it to destroy us. Releasing a powerful autonomous pentesting tool for free on the internet was incredibly naive. AI safety is not just about aligning models in a lab. It is about securing the autonomous agents we build on top of them before they tear down our financial systems.
Original reporting: Wall Street Journal. This is my rewrite and opinion.