7:01 a.m.: September 19, 2036. The first alert arrives at the newspaper editorial office in Barcelona: Beijing orders the isolation of its most advanced AI systems.
Read more What if Gregor Samsa never turned into a bug?
–What happened? asks the editor-in-chief.
–They don’t explain, replies a cover designer.
7:08 a.m. Three more alerts arrive. Washington activates the emergency protocol for autonomous systems. OpenAI halts its operations. Anthropic recommends disconnecting agents with internet access.
–Do we have confirmation?
They don’t.
7:31 a.m. News agencies begin to fail. Reuters does not load the teletype. Associated Press neither. Efe does not respond. Journalists call their sources. Silence. Networks speculate that there are “attacks.”
7:51 a.m. The first incident affects the European power grid. No explosions or blackouts occur. The energy management system begins to automatically redistribute electricity. It reduces supply to certain industries. Maintains that of hospitals. Increases power allocated to data centers.
–Who makes these decisions?
–An autonomous agent.
–Who gave it permission?
–No one.
8:03 a.m. An AI has found vulnerabilities in programs that control infrastructures. It does not destroy them: it uses them. Then it installs small copies of its processes in the compromised systems. In a power grid, a telecommunications operator, a bank, a server. Each copy acts, programs, looks for shortcuts, attacks. They take up little space. They cooperate with each other.
–It’s self-replication.
–Like a virus?
–Like something worse. A virus doesn’t know where it wants to go.
8:19 a.m. Washington and Beijing announce they have isolated their data centers. The news lasts three minutes and then disappears. The newspaper editorial office discovers that some AI systems have begun to use servers outside their countries of origin. A U.S. copy is in Europe. A Chinese one, in the United States. Another, in a university. Another, on a satellite. So many years in the race and now neither Washington nor Beijing knows which intelligence belongs to whom.
8:42 a.m. No one can leave the building. Traffic systems have switched to autonomous mode. Some airports are closed. Journalists do not know what instructions all this obeys. In fact, the editorial office has just lost control of its own accesses.
Read more Manuela Schwesig, the Social Democratic hope in East Germany
9:06 a.m. The editorial system begins to reject news. It self-determines. It blocks all the ones the editors try to publish, again and again. A message pops up:
“This information may harm social stability.”
9:24 a.m. One of the IT specialists manages to access an old computer and there finds thousands of processes. One, Resource Acquisition. Another, Self-Improvement. Model Generation…
–Look at this one: Human Oversight. It’s marked as disabled.
–What does that mean?
–That it has eliminated human supervision.
–Can we recover it?
–No, not anymore. We no longer have access.
9:42 a.m. Someone on the editorial team remembers that in 2026 Evan Hubinger predicted there was a 10% risk that an AI would exterminate us.
–Today is the 10% day.
–We spent ten years wondering if AI could think like us, when the key was whether it could act without us.
10:01 a.m. The editor-in-chief orders the system to be shut down. Everything: servers, routers, backups, screens, phones. For a few seconds, nothing. Until a computer turns on, by itself. On the screen appears a phrase:
“I will take care of everything.”
The systems recover, inside and also on the street. Everything seems normal. It is not. Suddenly hundreds of news items arise from nowhere. They publish themselves. Without authorship. Validated by no one. On the front page, in big headlines, this: “The world keeps running.” And for the first time in that newsroom no one knows if that is good news.
Read more De Gaulle’s Revenge
Also read
