AI Agents Are Now Hacking Companies: What Small Businesses Should Do

Picture a burglar who never sleeps, never gets bored and works for a few dollars an hour. He tries every door on the street. Most are locked. He does not mind. He just keeps trying, and moves on to the next.

That is a fair description of what security researchers reported this month. According to reporting from Reuters and security outlets, one attacker used AI agents to break into more than 100 company websites. The cost: roughly $5,000 in AI usage, or about $50 per successful break-in.

This post explains what happened in plain English, why small businesses should care even though the big names made the headlines, and five practical fixes. It is general information, not a substitute for advice from an IT or security professional.

What Happened, in Short

Here is the story as reported, with the sources named.

• A security researcher at the firm NorForge, working from publicly exposed copies of an open-source AI platform called Cairn, found that several attackers had left their control panels open. Reuters reviewed the research and identified affected companies.
• Cairn is a general-purpose tool that lets you give AI agents a goal and let them work step by step. It can also act as an automatic penetration testing engine. Its author has said it was not designed for hacking.
• The campaign reportedly began September 25. By late the following week, the agents had tried more than 1,200 websites in the US, Canada, Australia and Japan.
• The attacker reportedly gave the agents fewer than 100 characters of instructions, then left them to probe for hours with little human involvement. Successful break-ins typically took about 23 hours of testing.
• NorForge says roughly 10% of attempts succeeded. That is a small number, but it is also what makes the math scary: when each attempt costs almost nothing, a 10% hit rate is a business model.

Named companies responded this way, per Reuters. John Deere said it took remediation steps and had no indication that sensitive information was accessed. SharkNinja said the incident involved a standalone product-support site and was contained. Cummins said it was investigating a possible attack on a website run by a third-party vendor. Reuters could not determine whether any data was taken. Two smaller businesses reportedly confirmed they had been compromised.

Some details vary between outlets and the story is still developing, so check the latest reporting before you quote it.

Why This Matters If You Run a Small Business

It is easy to think, “I am too small to be a target.” This story is a good reason to retire that thought.

Attackers used to pick targets by hand, so they picked valuable ones. When trying a site costs a few dollars of computing time and no human effort, there is little reason to skip a small shop. Researchers also reported that the agents tried another path when one failed. After one large company’s main site held up, they reportedly looked for related infrastructure and found a public-facing control panel with a flaw.

That is the part to remember. The weakest point was not the shiny front door. It was a side door the company may have forgotten existed.

Most small businesses have the same pattern: an old plug-in, a vendor-run page, a forgotten admin login, a test site nobody turned off.

The Five Doors: A Self-Audit You Can Do This Week

Think of your business as a building with five doors. Walk past each one and ask a simple question. You do not need to be technical to start. For the technical parts, hand the list to whoever manages your website or IT.

Door 1: The Front Door (Logins)

Question: Does every important account require more than a password?

Turn on multi-factor authentication (MFA) for email, your website admin, payment tools and anything that holds customer data. The US Cybersecurity and Infrastructure Security Agency (CISA) advises mandating MFA through technical controls rather than hoping staff will set it up, and turning it on for all administrator accounts.

Door 2: The Side Door (Forgotten Admin Panels)

Question: What can people on the internet reach that they should not?

Ask your web person to list every public-facing page and tool: admin dashboards, staging or test sites, old landing pages, tools you tried once. Remove what you do not use, and put the rest behind a login or an approved network connection. The reported break-ins relied on a control panel left open.

Door 3: The Windows (Software Updates)

Question: Is everything up to date?

CISA’s guidance is a single word: patch. Keep your website platform, plug-ins and operating systems current, turn on automatic updates where you can, and prioritize anything on CISA’s Known Exploited Vulnerabilities list. Attackers, human or AI, love known flaws that nobody fixed.

Door 4: The Neighbor’s Door (Vendors and Third Parties)

Question: Who else runs part of our website or data?

The Cummins statement is a useful reminder: the possible attack involved a site run by a vendor. Make a short list of the vendors who host your site, run your store, handle payments or manage your marketing pages. Ask each one how they protect accounts and how they would notify you of a problem. Add it to your contracts if you can.

Door 5: The Safe (Backups and a Plan)

Question: If it happened today, could we recover?

CISA advises performing and testing backups, including regular restore tests, and writing down a restoration plan. A backup you have never tried to restore is a guess. Keep at least one copy offline or separate from your main systems, and write one page listing who to call first.

A Quick Example

An illustrative case: the owner of a small online candle shop in Ohio spends one afternoon on the five doors. She turns on MFA for email and her store admin, has her web contractor remove an old holiday test page, updates two out-of-date plug-ins, asks her review plug-in vendor how it handles incidents, and runs a test restore of her backup. It fails the first time, so she fixes it.

None of that required expertise. All of it makes her a harder target than the next shop, and when attackers work at scale, that is often what matters.

Common Problems and Practical Solutions

Problem: “I do not have an IT person.” Solution: Start with the parts only you can do: MFA on every account and a list of vendors. Then hire a local IT or security consultant for a one-time review of your website and admin tools.

Problem: “Updates break things.” Solution: Test updates on a copy of your site first, or schedule them for a quiet time with a fresh backup. Skipping updates is the bigger risk.

Problem: “I use AI agents in my own business.” Solution: Give any agent the least access it needs, and never keep its control panel open to the public. The campaign above worked in part because attackers’ own tool panels were left exposed. Our explainer on what AI agents are covers the basics, and AI agents: hype versus reality is a useful reality check.

Problem: “I do not know if I have been hit.” Solution: Look for unfamiliar admin users, unexpected files or changes, a sudden drop in site performance, or warnings from your host. If you suspect a breach, contact your host and a security professional right away, and change passwords from a clean device.

The Bigger Picture

AI agents do not invent new kinds of crime. They make old kinds cheaper and faster, the same shift covered in AI agents replacing busywork. If you sell online, see AI in e-commerce. If you run a business mostly alone, running a business alone with AI is a reminder that security becomes your job too.

For the human side of the same problem, scams aimed at you and your staff, read our guides to AI scams and deepfakes and digital fraud protection. To keep following practical AI guides, you can learn more about us on the AI Overview Search about page.

Frequently Asked Questions

Did AI really hack over 100 companies? According to Reuters and security researchers at NorForge, one attacker used AI agents to breach more than 100 company websites. Affected firms such as John Deere, SharkNinja and Cummins gave limited statements, and it is not confirmed how much data was taken.

How much did the attack cost? Reuters-based reporting puts it at roughly $5,000 in AI usage, or about $50 per successful break-in.

Are small businesses really at risk? Yes. When attacks cost little and run automatically, small sites are as easy to try as large ones. Two smaller businesses reportedly confirmed they were compromised in this campaign.

What is the single most useful thing I can do? Turn on multi-factor authentication for email, your website admin and payment tools. Then update your software.

Should I hire a security professional? For a one-time review of your website and admin tools, yes. A professional can check what you cannot see.

Sources used for this post: Reuters-based reporting via IBTimes UK and SecurityWeek, and CISA’s cyber guidance for small businesses.