One Hacker Used AI Agents Against 1,200 Sites. What Actually Worked?

A researcher traced more than 1,200 attempted intrusions to a hacker using AI agents. More than 100 sites were reportedly compromised. The numbers show the scale of the campaign, but attempts and successful breaches are not the same thing.

This content is blocked because it would connect to YouTube.

What the researcher reported

According to Reuters reporting by Raphael Satter, Snorre Fagerland of cybersecurity firm NorForge traced a campaign involving more than 1,200 attempted intrusions and more than 100 reportedly compromised sites. Those are separate measures: a target or attempt should not be described as a confirmed breach.

What the cost estimate does—and does not—show

Fagerland estimated that the attacker spent about $5,000 on AI-model usage, or roughly $50 per reported successful compromise. These are the researcher’s estimates, not a verified invoice or a universal price of carrying out a cyberattack. The case illustrates how AI services may lower the cost of scaling a campaign, without proving that every attempted intrusion worked.

What Reuters could verify

Reuters independently identified some affected organizations, while Deere, SharkNinja and Cummins gave different accounts of what happened. The extent of any data extraction could not be determined from the reporting. The story does not establish that the AI providers endorsed or knew about the misuse of their systems.

Why defenders should pay attention

The practical lesson is to treat AI-assisted intrusion attempts as a monitoring and response challenge, not as proof that every AI-driven attack succeeds. Organizations can review access controls, logging and incident-response readiness without sharing operational attack instructions. Watch the Daily AI Chat Short or listen to the full podcast episode. Source: Raphael Satter, Reuters, October 19, 2026.

Save to list

Scroll to Top