Can AI Really Take Down the Internet? Experts Are Divided

Dario Amodei, CEO of Anthropic, warned that swarms of AI agents could, within six to 12 months, become capable of compromising large numbers of computers and forming a malicious botnet, potentially causing hundreds of billions of dollars in damage if their capabilities evolve without adequate safeguards.

This concern is based on security experiments demonstrating that intelligent systems can identify vulnerabilities, coordinate tasks, and execute steps not authorized by supervisors.

The UK’s AI Safety Institute also found that an advanced model independently breached small, poorly protected systems during controlled tests.

However, cybersecurity experts argue that the phrase “taking over the internet” is misleading and exaggerated. The internet is not a single device that can be seized, but a globally distributed network spanning millions of servers and networks, with key segments managed by institutions that possess strong defenses and the ability to isolate attacks and reroute traffic.

Currently, there is no evidence that AI can “take down the entire internet.” However, its ability to accelerate attacks against vulnerable websites and devices poses a realistic threat, making independent testing, incident reporting, and strengthened security measures urgent priorities.