The Antivirus Racket Comes for the Agentic Internet

Welcome to Black October

The agentic internet just got its first protection racket, and nobody had to plan it.

This month the security world gave itself a name for a very bad stretch: Black October. Two critical flaws in PHP, the language behind roughly 70% of websites, went public on October 21 under the name PHP2Shell. Cloudflare saw hundreds of thousands of exploit attempts within about two hours of the news spreading. Governments took websites offline as a precaution.

The pile-up didn’t stop there. LiteLLM, an open-source gateway that developers use to reach AI models, had its release pipeline poisoned. An Indian certificate authority, eMudhra, turned out to have private keys weak enough to break, with one reportedly equal to the number 4. Quest Diagnostics confirmed a breach of patient data. And US banks reported something that should be the headline: unauthorized third-party AI agents operating inside customer accounts.

Here’s the detail that matters most. The PHP flaws weren’t found by a lone hacker in a basement. Researchers at Google’s Wiz and the AI firm Calif found them in August using advanced models from OpenAI and Google. The machines found the holes. Now everyone needs machines to plug them.

We’ve seen this movie before

If you owned a PC in the 1990s, you remember how antivirus worked. You bought a box of Norton or McAfee. A year later it nagged you to renew, because new viruses had come out and your old definitions were stale. The threat never ended, so the bill never ended either.

For years a rumor went around that the antivirus companies were writing the viruses themselves. There was never real evidence for it, and it wasn’t needed. The business model worked fine without a conspiracy. All it took was a permanent threat and a company standing between you and it, charging rent.

That’s the template. Hold onto it, because the AI version is about to get much better at the one thing the old racket only pretended to do.

The sword and the shield, same store

The antivirus rumor was false. The AI version is just how the market is built.

The same few labs that make the models finding vulnerabilities also sell, or will sell, the models that scan for them and patch them. A frontier model that can find a twenty-year-old flaw in PHP works for whoever is paying. Attack and defense run on the same engine, billed per token.

Look at the advice after Black October. Former CISA director Chris Krebs told smaller companies to scan for vulnerabilities constantly, patch, or take legacy systems offline. Constantly is the key word. Nobody scans constantly by hand. You scan constantly by paying for an AI that never sleeps, and you keep paying, because the attacker’s AI doesn’t sleep either.

That isn’t an arms race in the Cold War sense, where two sides spend until one goes broke. It’s a toll booth with a war painted on it. The fight is the product, so nobody selling it has a reason to end it.

Agents at the gate

Hacking is only half of it. The bigger market is defending a company’s whole presence online from other people’s agents.

That’s already starting under duller names. Cloudflare started blocking AI crawlers by default last year and floated charging them per crawl. A whole “generative engine optimization” trade has popped up to get brands recommended by chatbots instead of Google. So companies are paying to keep agents out and paying to get favored by them, sometimes at the same moment.

Now add the bank story. If unauthorized agents can operate inside customer accounts, every company with a login page needs its own agent standing guard, telling friendly agents from hostile ones in real time. Your shopping agent haggles with the store’s pricing agent while a scraper agent tries to undercut both. Every one of them runs on a model rented from the same handful of companies.

In the antivirus era, at least the virus writers and the antivirus vendors were different people. In the agent era, the referee, both teams, and the stadium might all bill the same parent company.

A fair caveat

To be fair, Black October is partly a story about a name. Some reporting points out that much of the month’s surge is disclosures landing together. Several breaches started back in 2025 or earlier this year and only came to light now. Giving a month a catchy label makes scattered events feel like a single wave, which conveniently helps whoever is selling the fix.

And the scariest single failure wasn’t AI at all. A certificate authority with a private key you could guess is ordinary human negligence. AI didn’t create the rot in the internet’s foundations. It just made that rot very expensive to ignore, very fast.

Renewal notice

So yes, companies will pay AI companies to defend their online presence from agents, and the arms race won’t end, because it was never meant to. That’s the antivirus deal. The difference is scale and conflict of interest. Norton never sold the viruses. The labs sell the capability on both sides of the fight, and the meter keeps running either way.

The question for the rest of us is whether our own agents stay ours. Once every nudge, every block, and every defense can be bought, does your agent work for you, or for whoever paid its maker the most this quarter?

Your subscription is about to renew. It always is.

Sources

Author: Shelton Bumgarner

I am the Editor & Publisher of The Trumplandia Report

Leave a Reply