Traditional hacking is labor-intensive: a researcher probes a site, writes exploit code, and repeats. Automated exploitation delegates that loop — probe, analyze, attempt, retry — to software. Security teams have used such tools for years, but AI agents add something new: they can improvise, chaining unexpected steps when a standard approach fails.

A Reuters report described a Chinese-speaking hacker who gave AI agents fewer than 100 characters of instructions and watched them target over 1,200 websites, successfully breaching more than 100 companies — including Deere and SharkNinja — for an estimated $5,000 total, roughly $50 per breach.

What's new isn't that hacking got automated; it's that the automation can now adapt. When one approach hit a dead end, the agents searched for related infrastructure on their own. That kind of improvisation previously required a skilled human in the loop.