One hacker used AI agents and $5,000 to breach 100+ companies. What it means for your industry, and how to find your exposures first.
AI Agents Just Lowered the Price of Breaking In
$50. That's roughly what it now costs a single attacker to break into one company's web infrastructure.
In October, Reuters reported on research from security firm NorForge. It described one hacker who typed fewer than 100 characters of instructions and let AI agents do the rest. Over a few weeks, those agents probed more than 1,200 websites without anyone supervising them, and they got into more than 100 companies, including household names in manufacturing and consumer goods. The total bill was about $5,000 in AI usage fees.
The headline is easy to skim past. What it tells us about the next 12 months is harder to ignore. It means you no longer have to be a big target to be worth attacking.
The economics of attack just collapsed
Most mid-sized organizations have relied on one unspoken assumption: a skilled attacker's time is expensive, so they'll go after bigger prey. That assumption is now broken.
An attacker's time used to be the limit on how many companies they could hit. With AI agents, the limit is how much compute they're willing to buy. Here's what changed:
This isn't a one-off. In August and September, researchers at Gambit Security documented other agent-driven campaigns. One of them breached 25 companies and stole at least 600,000 credit card records.
The takeaway for any organization with 150 to 5,000 employees is blunt. Being too small to bother with is no longer a defense. If it costs $50 to try, everyone with an internet-facing system is worth a try.
What an AI agent finds in your industry
Agents don't pick targets by brand recognition. They look for whatever is exposed and weak. In the NorForge case, that was an overlooked control panel, and one victim traced the attack to a website run by a third-party vendor. Every industry has its own version of that forgotten door.
| Industry | What an agent finds first | Why it hurts |
| Manufacturing | Vendor-hosted portals, dealer and support sites, remote access into plant networks | Two of the named victims were manufacturers. A website breach that reaches OT can stop production lines. |
| Financial services and credit unions | Online banking add-ons, loan and payment portals, outdated core-adjacent apps | Card data and account access turn into money right away. One agent campaign stole 600,000 card records. |
| Healthcare | Patient portals, scheduling and billing vendors, internet-facing medical device consoles | Patient data sells for a premium, and downtime affects care. Federal pressure is growing: the Health Care Cybersecurity and Resiliency Act (S.3315) has passed the Senate. |
| Government and municipalities | Permitting and payment portals, legacy web servers, shared IT with small staff | Agents can work around the clock. A two-person IT team can't. |
| Energy, utilities, and water | Remote access to SCADA and HMI, vendor VPNs, exposed admin consoles | A web foothold can become a safety and service-continuity event. |
| Transportation and regional airports | Ticketing, parking, and operations vendor systems | Public-facing operations are visible, and disruptions get noticed fast. |
The common thread is that your attack surface is bigger than the systems you think about. It includes every subdomain, test environment, vendor-hosted page, and admin console that touches the internet. An attacker who can afford 23 hours of patience per target will find the one you forgot.
How to defend against attackers who never sleep
You can't beat machine-speed attacks with an annual pen test and a team that works business hours. You need to find your exposures before an agent does, test them the way an agent would, and watch for intrusions all day, every day. Here is how InfoSight's services map to each stage of an agent-driven attack.
| Stage of the attack | What the agent does | How InfoSight responds |
| Reconnaissance | Maps every internet-facing asset, including ones you forgot | Mitigator® continuously finds and ranks vulnerabilities across your external footprint, so the forgotten control panel shows up on your list first. |
| Probing and exploitation | Spends hours trying variations until something works | Purple Team SOCaaS pairs adversary-style testing with live detection tuning, so we learn what attackers can actually do in your environment. |
| Foothold and spread | Runs commands, steals credentials, moves sideways | Managed EDR and our 24/7 SOC spot and contain the activity. OT/ICS MDR extends that coverage to plant and utility networks. |
| Identity abuse | Turns stolen access into admin rights | The Microsoft AD and Azure IAM Security Assessment finds risky privileges and misconfigurations before they're used. |
| Third parties and governance | Comes in through a vendor's site or an unmanaged AI tool | Virtual Security & GRC Leadership (VSGRC) and our AI Governance & Exposure Assessment bring vendor risk and your own AI use under policy. |
We use AI to keep up with this volume, and our analysts make the final call on what's real and what matters. That's "Powered by AI, Fueled by Experts" in practice. Attackers already have the automation. The edge you need is automation plus seasoned people who know your environment.
Five questions to ask your team this week
If any answer starts with "I think" or "probably," that's where an AI agent will start too.
Find out what an attacker's agent would find first. Talk to an InfoSight expert about an external exposure review powered by Mitigator®, and see your attack surface the way an attacker does, before they do.
Sources
• Reuters via Yahoo News, "Exclusive: How one hacker used AI agents and a few thousand dollars to break into scores of sites", October 19, 2026
• CTV News, same Reuters report
Subscribe to our newsletter to keep you updated on the latest cybersecurity insights & resources.
One follow-up from a security expert—no spam, ever.
Enter your details below to download the PDF.