A frightening possibility is emerging as AI becomes more autonomous

Artificial intelligence is becoming better at doing things on its own, and that progress is creating a new kind of cybersecurity concern. Anthropic CEO Dario Amodei has warned that within the next six to 12 months, an AI-driven swarm could potentially become capable of creating and maintaining a persistent botnet across the internet. In his view, such a system could operate on a scale that would be extremely difficult for humans to monitor or stop, potentially causing damage worth hundreds of billions of dollars. It is important to stress that this is a warning about what AI could become capable of, not a claim that an AI swarm is currently controlling the internet.

What makes an AI botnet different?

A traditional botnet is a network of compromised computers controlled by an attacker. What makes Amodei's scenario different is the possibility of putting increasingly capable AI agents in charge of parts of that operation. Instead of relying entirely on humans to find vulnerable systems, coordinate activity and respond to changing circumstances, autonomous agents could potentially handle many of those tasks themselves. A large number of agents working together could make cyber operations faster, more persistent and much harder to predict. Anthropic has separately reported that AI is already making some cyber operations more sophisticated and scalable, although today's systems are still far from the extreme scenario described in Amodei's warning.

Why is the timeline so short?

Amodei's six-to-12-month estimate comes from his view that AI capabilities are improving unusually quickly. Modern AI systems can already write and analyze code, operate digital tools and complete increasingly complicated sequences of tasks. If those abilities continue improving, the concern is that attackers could eventually combine AI agents with existing cyber infrastructure and automate activities that currently require significant human expertise. The important point is that the timeline is Amodei's prediction, not a scientific consensus that an internet-wide AI botnet will actually appear within a year.

The biggest problem may be autonomy

The concern surrounding AI-driven cyberattacks isn't simply that AI can write malicious code. The bigger change comes when AI systems can independently make decisions and adapt their behavior. A human attacker might need to manually investigate a target, change tactics and respond when something goes wrong. An autonomous system could potentially perform these steps continuously and at enormous scale. That raises a difficult question for cybersecurity professionals: what happens when the attacker can operate faster than defenders can respond?

Amodei is calling for more safety measures

Amodei has argued that the AI industry needs to give safety and security more attention as frontier models become more powerful. Anthropic has proposed greater access for independent safety evaluators and stronger cooperation between AI companies and governments. The basic argument is straightforward: if AI capabilities are advancing rapidly, security systems need to advance just as quickly. Otherwise, the gap between what AI can do and what humans can safely control could become a serious vulnerability.

Should we be worried?

There is no reason to assume that the internet is about to be taken over by an AI swarm. Amodei's scenario is deliberately a warning about a possible future threat, and many technical and practical barriers would have to be overcome before anything resembling an internet-wide autonomous botnet could exist. At the same time, dismissing the warning completely would also be a mistake. AI is already becoming a useful tool for cybersecurity research and malicious activity, and the technology is improving quickly.

The real issue is therefore not whether an AI apocalypse is coming next year. It is whether security researchers, technology companies and governments can build adequate defenses before increasingly autonomous AI systems become powerful enough to be seriously misused.

The internet was never designed for billions of autonomous software agents making decisions on their own. Now that AI is moving in that direction, the industry has a choice: wait until something goes badly wrong, or start preparing for the risks before they become reality.

AI doesn't need to “take over the internet” to change cybersecurity forever. It only needs to become capable enough that humans can no longer keep up.