Menu
A wall of vintage televisions with dark screens, static and bright color test bars.

The Hacker Has AI Agents. The Algorithm Has Your Attention.

, , , , , , ,

A person with a computer, a pile of time, and a modest budget has always been able to cause trouble on the Internet. What changed is how much of the work that person has to do themselves. As of last week, the answer appears to be: not very much.

On October 20, Raymond Tec News covered research into AI agents breaching more than 100 websites. I want to take that story further, because the same broad change that lets software keep trying attacks without a human babysitting it also affects how information is generated, distributed, and believed.

This is part three of The Infrastructure We Trust. The first two articles looked at vulnerable systems and the things we depend on. This time the vulnerable system might be an ecommerce application. It might also be our own understanding of what’s happening in the world.

One attacker, more than a thousand attempts

On October 19, Reuters and security researchers at NorForge described a Chinese-speaking operator using Cairn, an open-source system for coordinating AI agents, to attack online stores and associated infrastructure. According to Reuters’ independent reporting, the campaign attempted intrusions at more than 1,200 sites and reached more than 100 company websites. Some of the sites belonged to large, familiar businesses; the reporting also included instances involving vendors or stand-alone support websites, rather than necessarily the company’s primary storefront.

The operator wasn’t issuing a detailed sequence of commands for each site. The AI agents performed reconnaissance, explored connected assets, searched for weaknesses, and kept trying. Successful break-ins could take around 23 hours, but that’s 23 hours of a system working while its operator did something else. NorForge estimated the AI-model costs at roughly $5,000 for the campaign, or around $50 per successful intrusion.

Language is not nationality, and nationality is not proof of government sponsorship. Calling the operator Chinese-speaking tells us something about the instructions the researchers saw. It doesn’t tell us who paid the bills, who directed the attacks, or whether a state was involved. Reuters also said it couldn’t establish how much data, if any, was extracted in all the cases. A confirmed compromise and a confirmed theft of customer records are not the same thing.

The thing that worries me is not that AI has invented an entirely new class of weakness. It’s that it makes the old ones cheaper to find. An abandoned admin panel, a forgotten support site, a stale credential, or one badly protected file store can receive the sustained attention of a tireless automated adversary. I wouldn’t hire a human tester to spend a full day guessing at every neglected server on the Internet. An agent doesn’t mind.

Turning on AI is the easy part

Deciding what an AI tool should be allowed to see, who should use it, what work it should perform, and what happens when it gets something wrong is the more interesting problem. Raymond Tec helps businesses connect and automate the tools they actually use without treating every new feature like a button that obviously needs to be switched on.

AI is useful to defenders, too

This isn’t an argument to ban AI from the server room. Google and Wiz reported using AI-assisted research in the discovery, analysis, and remediation of the critical PHP runtime vulnerabilities disclosed October 21. Defensive teams can use automation to review code, correlate logs, find forgotten services, and test configuration changes.

The difference is authority and verification. A defensive agent should operate with deliberately limited permissions, record what it changed, and have a human-reviewed recovery path. I don’t want a clever automated assistant deciding that the best way to clean a compromised production server is to delete something essential. We’ve already had enough fun with ordinary automation making bad assumptions.

For a business, this is where the practical work stays surprisingly familiar: find every public-facing application, keep dependencies current, remove unused services, protect administrative logins, watch for unexpected changes, and maintain independent backups. What AI changes is the urgency and volume of the tests your defenses may face.

Now let’s talk about AI and what we believe

I’ve been down a conspiracy-theory rabbit hole this week. With wars affecting energy prices, infrastructure disruptions in Europe and the United States, and widespread demonstrations, I wondered whether some of these events were being coordinated. Maybe by foreign intelligence services trying to destabilize other countries. Maybe by domestic power structures trying to benefit from the resulting chaos. It’s a tempting story because the timing of some events is genuinely unsettling.

Then I tried to prove it. I could document real attacks. In some cases I could document coordination within a particular campaign. I found precedents for state-sponsored sabotage, as well as private criminals and opportunists exploiting weak infrastructure. What I couldn’t find was the evidence joining French supermarket refrigeration, Ahold’s network intrusion, American water systems, and the protest movements into one operation.

I still think the questions are worth asking. But a good conspiracy theory is a hypothesis that should survive a hostile fact-check. And that’s precisely where AI-generated posts and social media can lead us astray: they can manufacture a coherent-looking story much faster than anyone can verify it.

Content & search

SEO isn’t a magic field hidden in your admin settings. Useful structure, understandable copy, good metadata, sensible internal links, and content worth landing on still do most of the work. Raymond Tec helps businesses improve the whole publishing system instead of chasing one green score.

A protest movement can acquire an audience before it acquires an organizer

France offers a particularly revealing example.

Ahead of the October 17 demonstrations, Le Monde traced viral calls for a Yellow Vest return on TikTok.

The newspaper identified approximately a dozen accounts created or renamed on the same day in September. Some circulated AI-generated posters promoting supposedly official protest dates and meeting places. Yet prominent veterans of the movement said the organizers were unclear, and some warned supporters against treating the posts as authoritative.

Here’s the part that complicates the easy foreign-agent explanation: France’s Viginum agency told Le Monde it had seen only minimal foreign amplification without meaningful reach. The account behavior may have been driven largely by people trying to acquire followers they could later monetize or sell. That’s a very ordinary motive sitting inside an extraordinary-looking political moment.

It’s possible for a platform to create the appearance of organization without an actual central organizer.

  • One account posts a graphic
  • Others copy it
  • The algorithm pushes it
  • People see the same claim repeated enough that it starts to feel established

Real political anger still exists. Real people still march. But the origin of the specific information they’re seeing can be difficult to untangle.

Yesterday’s fan page, today’s influence machine

Accounts can also change identity. Sometimes an owner rebrands or sells one. Sometimes someone steals it. An existing audience then encounters material from a completely different source while the old follower count and history appear to confer credibility.

This is not just a hypothetical tactic. A Harvard Kennedy School Misinformation Review investigation documented a music group’s Facebook page that was reportedly hacked, renamed, and repurposed to share AI-generated nature images. It rapidly gained hundreds of thousands of followers. Researchers also described other repurposed pages formerly associated with a church and unrelated businesses.

The Harvard cases demonstrate account hijacking and audience repurposing; they don’t prove that those accounts later became political propaganda channels or that the French protest accounts were hacked. But you can see why the technique matters. Change the theme of a page with a large existing following, and the distribution network already exists. Repeat that across many accounts, and the potential reach can become substantial without building a new audience from scratch.

The rules around technology matter too

Platforms, privacy, speech, competition, surveillance, copyright, and regulation increasingly determine what technology companies can build and what the rest of us have to live with. Browse more Raymond Tec News for practical coverage of technology policy and digital rights.

The No Kings suspension shows another kind of platform fragility

During the October 17 No Kings demonstrations in the United States, the coalition’s Instagram account was temporarily suspended. Reuters reported Meta’s explanation: automated security systems mistakenly classified unusual activity associated with a Claude-based tool as bot-like behavior. The coalition said it lost its ability to communicate through the account for hours and demanded a fuller explanation.

There is no verified evidence that a government directed the suspension. Nevertheless, the practical consequence was real: a movement’s ability to communicate depended on rules, algorithms, and account-recovery decisions controlled by a private platform. This is why media literacy isn’t only about spotting fake pictures. It’s also about understanding who controls the channels through which pictures and messages reach us.

Before sharing a dramatic claim, I now want to know: where did it originate, who first documented it, can I find a primary source, and does independent reporting agree on what actually happened? If an account has changed its name, subject, or ownership, that history matters. Screenshots without provenance and identical captions repeated a hundred times aren’t a substitute for evidence.

The machine can write the story, make the poster, and keep pushing it. The responsibility to decide what’s true hasn’t gone anywhere. It’s now even more work, proving that real human journalism will never die in the wake of AI.

News tells you what happened. Sometimes the interesting part is what happens when you actually have to work with this stuff. Raymond Tec Insights collects longer field notes, lessons, opinions, technical decisions, and stories from real projects.

Sources and further reading

Featured photo: Marshal Yung via Pexels, used under the Pexels License. The image is illustrative and does not depict an actual cyberattack or coordinated influence operation.

Related Posts

Leave a Reply

Only people in my network can comment.