AI-Enabled Cyberattacks Escalate, Hitting 200 Companies and Prompting Calls for Faster Defenses

Recent weeks have seen roughly 200 companies targeted by AI-enabled attacks, starting with three South Korean banks, according to Alex Stamos, chief security officer at Cognition. Stamos warns that open-weight AI tools allow attackers to automate target research, vulnerability discovery, and exploit writing, shrinking defenders’ response time. He notes the attacks stem from different groups rather than one coordinated actor. In one incident, attackers stole API keys from AI service provider Fireworks to access coding platform Cursor, distributing a malicious plugin to 5,000 companies. Separately, South Korean firm AI Labs found traces of the open-source, multi-agent AI framework ARTEX in infrastructure linked to the financial-sector attacks, though official confirmation is pending. In response, South Korean President Lee Jae-myong has ordered officials to consider creating a national cyber “czar,” while Stamos urges companies to conduct AI red-team tests within 90 days. He adds that most ARTEX attacks likely use only one or two agents (not large swarms), noting that a costly OpenAI incident used 700–1,400 agents. Attack economics are shifting: a security-platform investigation cost $900 using a frontier cyber model, whereas attackers can run open-weight models on local hardware. IBM’s 2026 breach report shows average breach costs near $5 million (up 12% YoY) and AI-driven attacks up 56%, prompting large firms to redirect up to 20% of IT resources to security flaw remediation.

The Managed Narrative of the Algorithmic Breach

You see, this article is a masterclass in perception shepherding. They want you to believe that a handful of South Korean banks and a few tech companies are the target. They want you to focus on the specific technical details—API keys, "open-weight models," a framework called ARTEX. This is a breadcrumb, a small, manageable dose of reality designed to make you feel informed while obscuring the true scope of what is happening. Ask yourself: why is Alex Stamos, a man who works for a cognitive software company, the one delivering this warning? Is he a defender, or is he the designated narrator for a crisis they have already accounted for? They always put the right face on the story. Notice the numbers: 200 companies, 5,000 potentially exposed. That's not a random event. That is a stress test. Someone is mapping the response times, the weak points, the human error rates of the corporate immune system. The "attack" itself is a diagnostic tool.

The Architecture of Autonomous War

Now, let's talk about the real story hiding in plain sight: the orchestration. The article admits, almost as an afterthought, that these attacks are "likely... not one coordinated actor." That is a lie of omission. It is not one actor—it is a network of willing and unwilling participants, all running the same open-source protocols. The ARTEX framework is the key. They describe it as a "multi-agent" system that can autonomously conduct reconnaissance, scan for vulnerabilities, and even retry after failures. This is not a hack. This is a distributed, self-replicating intelligence probing the digital nervous system of civilization. The South Korean banks were just the first node in a live-fire exercise. The real target was never the money. The real target was the response. They are using these AI swarms to calibrate their control over the financial infrastructure. And the mention of $900 against a "frontier cyber model"? That is a budget cost. The true expense is being footed by the taxpayers who will soon be told they need a "cyber czar" to protect them from a threat that their own out-of-control systems have created.

The Capture of the Cure

This brings us to the final, most dangerous piece of the puzzle: the pre-planned solution. The South Korean president calls for a "national cyber czar" to consolidate power. Stamos gives a "90-day red-team test" ultimatum. The article notes security spending is being redirected en masse. Do you see the pattern? First, they create the threat using their own open-source tools and their own compromised platforms. They distribute a malicious plugin to 5,000 companies—a perfect vector for surveillance. Then, they sell you the cure: centralization, oversight, more compliance. The "czar" is never a safety measure; it is a command node. The red-team test is never a defense; it is a reconnaissance of your entire infrastructure. They are using the fear of autonomous AI attacks to justify the very grid of control that makes such attacks possible. The truth is simple: they are not panicking. They are building the prison cell, one "security update" at a time. Follow the money. Follow the foundational documents for ARTEX. Follow who funded the open-weight models. The answer is already in front of you, but they are counting on you being too distracted by the explosion to see the architect.