Agentic Ransomware

There is a type of malicious software that makes all files on a computer unreadable and then demands money for their release. This principle is called extortion software, or ransomware. Until now, such a program was a rigid sequence of commands: the attacker specifies in advance exactly what should happen. With agentic ransomware, an AI program is built in that looks around the foreign environment on its own and continuously decides what the next step should be. It can therefore react to surprises instead of simply stopping. Security researchers have been observing such attacks since 2025, and the term has since appeared regularly in reports from major security firms.

Why extortion becomes cheaper as a result

A good attack on a corporate network used to be manual labor. People had to click their way through foreign systems, find passwords, and figure out where the important data was located. That costs time and requires expertise. With agentic ransomware, it is precisely this manual labor that the software takes over.

This significantly lowers the entry barrier for criminals. Those who previously needed an experienced team may now get by with a rented tool. At the same time, many victims can be attacked in parallel, since no human has to handle each case individually anymore. Experts therefore speak of scaling: the same amount of work suddenly hits a hundred times more targets.

For companies and hospitals, this is a real risk. A single successful attack can paralyze operations for weeks. Insurers and regulators are watching the development closely, because the damage sums keep rising.

What the AI part takes over in the attack

At its core is a language model, that is, a program that can generate text and commands. It is given a goal, for example: find all customer data and encrypt it. It then works in a loop. It checks what is present on the system, considers a next step, executes it, and reads the result.

Such independently operating programs are called agents. The agent can search through folders, read out password files, or work its way toward other computers on the network. If an attempt leads nowhere, it tries a different route. A rigid piece of malware would simply have stopped at that point.

A common misconception is that the AI invents these attacks itself. It does not. A human sets the goal, builds the tool, and collects the ransom. What is new is only how much of the intermediate work the machine handles. Technically, too, much remains familiar: the encryption itself is ordinary mathematics, not an AI trick.

Where the term appears in the news

You mainly read it in reports from security firms and in news about attacks on clinics, city administrations, or logistics companies. Providers of AI models also comment on it. They report when accounts were suspended because someone tried to misuse their systems for such attacks.

For investors, the topic is interesting because it affects two industries. IT security firms advertise defenses that likewise rely on AI. Insurers are recalculating their prices for cyber policies. Both show up in quarterly reports and market forecasts.

In your own daily life, the issue affects you indirectly. When a school, a doctor’s office, or an online shop is suddenly offline, such an attack is often behind it. The most important protection is unspectacular: regular backups on separate storage media, up-to-date updates, and caution with email attachments.