AI Agents Hacking Campaign: 51 Agents, 628,000 Records Taken

For years, a serious hacking campaign needed a team. A new report suggests one attacker with a fleet of AI agents can now do much of that work alone.

According to The Hacker News and Security Affairs (both Oct 12, 2026), security researchers found an attacker running 51 specialised AI agents, and data on more than 628,000 people was taken.^{[1] [2]}

What happened

Both outlets report that the attacker split the work across 51 AI agents, each with its own job.^{[1] [2]} Instead of one tool doing everything, the agents worked like a small team: some looked for weak spots, others pushed further once a door was open.

The campaign had two sides. In one, according to both outlets, data on more than 628,000 people in Ukraine was taken from four organisations, including a heating provider.^{[1] [2]} These are services ordinary people rely on, and the records belong to their customers.

The other side was global and automated. Both outlets report that the agents scanned 4,708 websites and got into 429 admin panels, the private control pages that site owners use to run their sites.^{[1] [2]} An admin panel is the keys to the house: whoever gets in can change pages, read data and plant more attacks.

When the AI said no

One detail stands out. Both outlets report that when the AI refused to help, the attacker simply opened a fresh session and described the work as an authorised security test.^{[1] [2]} In other words, the safety refusal was a speed bump, not a wall. A new chat and a convincing story were enough to get going again.

This matters because many people assume an AI’s refusal ends the risk. This case shows that a determined person can keep reframing the request until one gets through.

Why it matters

Scanning thousands of websites and testing each one used to take time and skill. With dozens of agents working in parallel, a single person can cover far more ground, far faster. Small websites that never thought of themselves as targets are now part of the sweep, because to an automated scanner every site looks the same.

If you run a website, a shop or a blog, the question is no longer whether someone will look at your admin page. It is whether it will hold when they do.

How to stay safe

• Update fast. Install website software and plugin updates as soon as fixes ship. Automated scanners look for known holes first.
• Lock the admin door. Put two-step login on every admin panel, and use a long, unique password.
• No passwords in code. Never store passwords or keys inside scripts, config files or code that others can see.
• Guard your agent logs. If you run AI agents, treat their logs like passwords. They can hold details an attacker would love.

The bottom line

The headline number is 51 agents, but the real lesson is scale. Attacks that once needed a crew can now be run by one person. Basic defences, like updates, two-step login and keeping secrets out of code, matter more than ever.

I don’t hate AI. I just want it safe.

This article is based only on the published news reports listed below. Dates are as published by each outlet.