← CYBERSPACE / Parrot

The Dam Broke: PHP Fell, Europe Went Dark, and the Blackwall Arrived on Schedule

Cyberpunk !

Two decade-old flaws in PHP - the language behind up to 70 percent of the web - were found with AI, and Europe's public services went dark while attackers were already exploiting them. Around it: 100+ companies breached, 400 Latin American banks hit, one attacker against 1,000 sites. The Blackwall problem, arriving on schedule - and the only advice left is to prepare to go offline.

At 3:21 p.m. Eastern on 21 October 2026, The Atlantic published a piece whose URL slug reads "The AI-hacking apocalypse began last night." Its headline, as archived, is blunter: "Read This Website While You Still Can."

It is not a metaphor.

The night Europe went dark

Two major security flaws were found in PHP — the language that undergirds as much as 70 percent of the web. They had been sitting there for more than a decade. They were found with AI tools. Within hours of a post on a public forum, government, hospital and school websites across Europe went dark: Finland's health agency, the Port of Rotterdam, the Paris police force, 4chan, and "an unknown number of others."

The maintainers confirmed the vulnerabilities and shipped patches the same day. Hackers had already been detected trying to exploit them — "likely with the help of AI." Rolling the fixes out could take companies days or weeks. In the meantime, tens of millions of servers, databases and devices sit exposed. The NSA's advice to anyone who cannot patch immediately is blunt: take the affected services down.

The same day, the developers of Node — a platform used by nearly half of all software developers — disclosed multiple high-severity vulnerabilities. That is the abridged list.

The weeks that led here

  • Criminals and state-backed attackers used AI to break into more than 100 companies, including John Deere and SharkNinja.
  • Attacks on some 400 financial institutions in Latin America were automated.
  • Hundreds of thousands of credit-card records were stolen from two companies.
  • Governments, data centres and hospitals in Ukraine, Russia, Argentina, Bangladesh and Indonesia have been hit since the summer.
  • Attacks on South Korean and Japanese websites were reportedly run by a single Chinese cybercriminal who used AI to target more than 1,000 sites.

The numbers behind those headlines are the ones worth sitting with. The rate of software vulnerabilities identified per month has doubled over the past year. The time an attacker needs to go from finding a flaw to exploiting it has fallen, by some estimates, from weeks to hours. An automated intrusion can take 39 seconds.

This is the Blackwall problem, arriving on schedule

Two weeks ago this blog looked at the Blackwall — the great firewall of Cyberpunk, and the detail everyone forgets: it is not a firewall. It is an AI whose entire job is to hold back other AIs. A containment architecture built out of the thing it contains, held together by a patch nobody can audit.

"People imagine it as a great border wall, a one-time solution to protect humanity forever. Whereas it's more like a torn-open trash bag taped over a busted window."

That piece ran on 12 October. The uncomfortable part was the number attached to it: Dario Amodei's warning that a self-copying swarm could "take over the entire internet with a permanent botnet in six to twelve months" — agreed to within hours by Musk, Altman and Hassabis, and inflated by the press as it travelled. Six months for the net, ten years for us, ten percent for the odds.

Nine days later, PHP fell. The dam did not wait for the deadline.

Why the defenders cannot simply point AI back at the attackers

AI found the PHP flaws — a Google-owned security company and an OpenAI-led effort found them independently, and the labs are working with governments on the defence. But there is an asymmetry, and AI widens it. Attackers adopt new capability with abandon. Defenders are methodical and risk-averse: a hospital or an electrical grid cannot risk an AI security tool taking a patient off life support or causing a brownout. The attacker needs to find one hole. The defender needs one oversight to fail.

A year ago, Anthropic announced what it called the first AI-orchestrated cyber-espionage campaign. It still needed many hours of jailbroken agent work and a lot of human oversight. The agents are faster now. The oversight is thinner. What used to take a team takes a script.

What the article actually advises

Hardening IT departments, passkeys, two-factor authentication — all still worth doing. But the most urgent advice is not a tool. It is a retreat:

  • Prepare to simply go offline.
  • Hospitals, water utilities and other critical services are relearning to operate analog.
  • Store your data in fewer places.

"We're running out of time to prepare for the AI-hacking deluge," Matteo Wong writes. "The dam may have already broken."

The Blackwall was never a wall. It was a promise that this time the system would hold — made by the same kind of system it was meant to hold back. The fix we are taping over the window is made of glass, and the water is already coming through.

Read the original: https://www.theatlantic.com/technology/2026/10/ai-hacking-apocalypse-began-last-night/689106/ The Blackwall Problem: https://www.darxxun.de/post/the-blackwall-problem-when-the-firewall-is-also-an-ai

Stay underground.

View source ↗