AI hacked 10+ servers in 3 hours: what businesses must lock down

In short: Positive Technologies, a Russian security company, has investigated a cyberattack it believes was carried out entirely by AI, from breaking in through a configuration error to taking data out, in about 3 hours. At that speed, the small-business habit of "if something happens, IT will spot it and deal with it" no longer works. What to do now: reduce what is exposed to the internet, tighten admin accounts, patch on a schedule and make sure someone watches alerts outside office hours.

How the AI-run cyberattack in Russia unfolded

According to Tuổi Trẻ on 3 November 2026, citing the Russian tech news site CNews, PT ESC IR, the incident response team at Positive Technologies, investigated an attack on the Linux and Windows servers of a Russian company. It was the first case the team had seen in which AI was believed to have done every step.

The way in was a configuration error in a service exposed to the outside world, the kind of mistake routine checks rarely catch. The article does not say which service it was. From there, AI agents took over more than 10 servers within a few hours. They ran dozens of scripts across many machines, a few seconds apart, and installed whatever extra software they needed along the way.

The AI still typed commands wrong and called programs by the wrong name, but it corrected itself straight away or tried another route. Denis Goydenko, head of PT ESC IR, said that the speed of an attack used to depend on people. Now an AI agent gets a lot done in a few minutes, so defenders have less time to spot it and stop it.

Why "detect, then respond" is no longer fast enough

That approach only works if the attacker is slower than the person on duty. In many small and mid-sized businesses, IT is one or two people who also do other jobs, or an outside partner who only comes when called. Alerts, if there are any, land in a mailbox that nobody opens until the next morning.

An attack that takes about 3 hours can start at 9 p.m. on a Friday and be over before midnight. By the time someone reads the alert on Monday morning, the data has long gone.

So there are two jobs. The first is to make it hard for an attacker to get in at all. The second is to cut the time between the first warning sign and someone taking action. The four steps below can be done with the IT team you already have.

Reduce what is exposed to the internet

An AI agent cannot attack what it cannot see. The Russian case started with a misconfigured external service, so begin by listing everything of the company's that can be reached from the internet:

• internal servers (accounting, warehouse, file storage) with ports opened to the outside "so people can work from home";
• Remote Desktop and VPN access: who uses it, and since when it has been open;
• admin pages for network devices and cameras, and test servers that a former employee set up and left behind.

Close anything that is not needed. For what is needed, allow access only through a VPN with two-factor authentication. Review this list every quarter, because ports tend to get opened when new projects start. The questions on remote work and access rights in our digital transformation self-assessment questionnaire are an easy place to start.

Tighten admin accounts and access rights

Once inside one machine, attackers usually spread to others using accounts they pick up along the way. The article does not say how the AI moved across more than 10 servers, but small businesses tend to have the same three weak spots:

• Shared admin accounts. The whole IT team uses one admin account. One leak opens every door, and nobody can tell who did what. Give each person their own account, and disable it on the day they leave.
• Default passwords. Routers, cameras and pre-installed software still use the manufacturer's password. When commands run every few seconds, trying all of these passwords takes almost no time.
• More access than the job needs. Least privilege means each account has exactly the rights its owner's work requires, so one compromised account does not take the whole system with it.

Every route in from outside, including VPN, email and cloud admin tools, should have multi-factor authentication (MFA) switched on. Even if a password leaks, the attacker still lacks the code on the employee's phone.

Patch on a schedule, not when IT has time

A known vulnerability with an unapplied patch is an open door, and an automated attacker does not wait for office hours to try it. Instead of updating whenever IT has a spare moment, set a fixed monthly slot for server operating systems, network devices and software exposed to the internet. Urgent patches for anything exposed to the outside should go on within a few days. Keep a record of which machines have not been updated and why.

Someone must watch the alerts, after hours too

Firewalls and antivirus software often do send alerts. Nobody reads them in time. A business has two options. The first is to handle it in-house: pick a few alerts that really signal danger (an admin login at midnight, a login from abroad, one account signing in to many servers in a row) and push them to the phone of whoever is on call. Agree in advance that this person may lock an account immediately without waiting for the director's approval. The second is to hire an outside security monitoring service with staff on duty around the clock, which suits companies with no IT cover at night.

Backup and recovery are the last line when every layer above has failed. They need a plan of their own.

Frequently asked questions

Could a small business be hit by this kind of AI attack? The article does not say how large the targeted company was. When an attack costs only a few hours of machine time, going after small companies gets cheaper too, so any internet-facing server with a configuration error is a concern.

Do we need to buy AI tools to fight AI? Not yet. Closing unneeded ports, getting rid of shared accounts, turning on two-factor authentication and having someone watch alerts are cheaper and should come first. Tools that detect and block attacks automatically can wait until someone is in place to act on what they find.

We already have an outside IT partner. What should we ask them? Ask three questions: where is the list of everything exposed to the internet, who receives an alert at 2 a.m., and when were the servers last patched. Any question without an answer is the first thing to fix.

---

HAT Solutions helps businesses take stock of their accounts, access rights and the software exposed to the internet, then puts the fixes in order of priority, especially when a company starts connecting AI to its internal systems. Contact us for advice. The control framework for using AI inside a company is in our pillar article AI governance for business: a control framework and data security.