Breaking Into a Company Website Now Costs About $50

by Oct 23, 2026ai, Business, security, Technology0 comments

AI agents used to automate attacks against company websites at low cost

The short version

  • An operator pointed AI agents at more than 1,200 company websites and got into over 100 of them.
  • Total spend was around $5,000. Roughly $50 per successful break-in, about $5 per failure.
  • The flaws exploited were ordinary: SQL injection, unsafe file uploads, outdated components.
  • Being small no longer protects you, because nothing is being targeted in the way that word used to mean.

Reuters published an investigation this month into a campaign that began in late September. An operator working in Chinese pointed AI agents at more than 1,200 company websites and got into over a hundred. The figure worth sitting with is the cost: around $5,000 in total AI usage, which works out at roughly $50 per successful intrusion and five dollars per failure, at close to a one in ten success rate.

What that changes

For years the honest answer to “why would anyone target us” was that mostly nobody would. Targeted attacks took skilled human time, that time is expensive, and attackers concentrated where the payoff justified the effort. At five dollars a failed attempt, the arithmetic stops holding. Nobody has to decide you are worth attacking. They point the system at a list and come back later to see what opened.

A failed attempt costs about five dollars. At that price, nobody needs to form an opinion about whether you are worth it.

The flaws were nothing exotic

The agents were not finding novel vulnerabilities. They worked through the standard list: SQL injection, unsafe file uploads, login bypasses, insecure handling of stored data. One intrusion chained weak session storage with an unsafe PHP deserialization flaw and an outdated Zend Framework component. Another found an administrative panel exposed to the internet and ran commands through it. These are things a competent review catches, and they have been on every web security checklist for a decade.

Targets were mostly online stores across the US, Canada, Australia and Japan, including sites connected to Deere, SharkNinja and Cummins. Several involved a site somebody else built or still runs: a support portal, a third-party-managed page. Those rarely appear in anyone’s patching schedule, and they usually sit on the same domain as everything that matters.

What to do

Inventory everything you have facing the internet, not just the main website, because most businesses have more than they remember and the forgotten ones are reliably the vulnerable ones. Then keep public-facing applications and their plugins patched, which given the critical PHP flaws fixed this week is more pressing than usual. Get administrative panels behind an IP restriction, and look through your web directories for files nobody put there.

The uncomfortable part of this is not the AI. Automated agents succeeded roughly one time in ten against real companies, using weaknesses documented for years. The tooling got cheaper. The holes were already there.

What do you have facing the internet?

Including the microsite nobody has looked at since 2024. PTSI inventories your external footprint, finds what is exposed and unpatched, and closes it before an automated scan does it for you.

Book an external review

PTSI Editorial Team

Support Line: Phone: +1 646-535-HELP (4357) Email: helpdesk@progressny.com Support web: helpdesk.progressny.com