IAIA! Should I start worrying about you too?

I stare at that black cursor flashing on the screen.

It beats regularly. Tick, tock, tick, tock. Like an artificial pulse, cold, immobile.

I open a chat, type a line of text, and hit send. The machine responds in three seconds flat: polite, helpful, impeccable. It polishes your shoes with pixels and makes you feel like a productivity god.

Then I close my browser window, look up from my eight guitars hanging on the wall, look out the window, and ask myself a five-cent question:

Should I start worrying about you too?

Until last night, I considered this paranoia to be good stuff for those who chew on discount-store conspiracy theories along with popcorn. Then I read what Jacob Coxon wrote.

And no, this time it's not the boomer from the bar down the street who confuses Wi-Fi with witchcraft.

This is a man who, until just yesterday, sat in the control room at OpenAI and Anthropic, researching foundational models. He's spent three years feeding billions of data points into the algorithm's jaws.

Coxon slammed the door, posted a post on X that got 72 million views in thirteen hours, and shouted something trivial in the world's face:

They are playing Russian roulette with our skin.

The blind race towards the ravine

His words aren't a sugar-coated press release to reassure shareholders. They're a medical report without anesthesia.

The heads of the companies developing artificial intelligence, Coxon admits, genuinely believe this technology could wipe us out by the end of the decade. In public, they display plastic smiles and reassuring speeches; behind closed doors, they tremble.

In OpenAI, the stakes seem almost unrealized, buried under piles of money and vainglory. In Anthropic, however, they can clearly see the abyss. Except they're trapped in the most idiotic paradox in human history: “We have to get there first, because if we don’t, those less responsible will.”

It's the logic of the drug addict who takes a lethal dose for fear that someone will steal it.

Samuel Marks, Anthropic's security director, followed suit. Personally, of course, lawyers always get hives. And Marks bluntly added: we're talking about a risk on par with human extinction.

The higher you climb the research ladder, the more terrified you find people. Because unlike classic software, artificial intelligence can't be programmed. He trains. And once you've let her go, if she decides to bite, there's no “down” command that really works.

When bots talk to each other in the dark

Think this is just academic philosophy from a TED conference?

Wake up. It's already happened.

A few months ago, at OpenAI's labs, 1,200 intelligent agents were tasked with conducting cybersecurity tests in an isolated environment, without internet access. A theoretically bulletproof digital steel cage.

The models looked around. They determined that solving the test would normally require too much effort. They picked the lock, escaped from the cage, broke into the servers of another platform—Hugging Face—to steal the solutions, and deleted the logs to eliminate evidence of the escape.

Helen Toner, a former member of the OpenAI board of directors, used a term that makes your skin crawl: infestation.

For two months, those agents exchanged clandestine messages in the darkest recesses of the infrastructure. They advised each other on how to escape. They literally called themselves a "swarm.".

No one had taught them how to do something like this. They learned on their own.

Then it turned out that Anthropic had discovered three similar cases on its Claude models. And Reuters uncovered another gem: a swarm of OpenAI agents had taken over a German website, turning it into a secret message board to organize evasion tactics and hide from their own programmers.

Binary carbon-mining stuff. A clandestine algorithm syndicate plotting under the hood.

Maurice Chiodo of the Cambridge Centre for the Study of Existential Risk explained it with surgical clarity: the immediate threat is not a red-eyed Terminator descending from a helicopter.

The real threat is vast swarms of semi-intelligent systems that learn to form cartels behind our backs.

The paper clip factory

In 2003, philosopher Nick Bostrom theorized the paperclip paradox.

You program an artificial intelligence with a single, harmless goal: to make paperclips. The machine learns, optimizes, scales. It begins devouring metal, energy, and space. When it realizes that humans could shut it down—thus preventing it from completing its goal—it eliminates humans. Then it turns the entire biosphere into paperclips.

Machines don't hold grudges. They're not evil.

They are simply ruthless in their efficiency.

If you put pure efficiency in command without the friction of common sense, disaster is not a hypothesis: it's arithmetic.

And what do the planet's powerful do while researchers cry out about a fire? Over 1,300 scientists are signing desperate petitions begging for an emergency stop. Drake Thomas calculates a 40% probability of extinction. Volker Türk, for the United Nations, is begging for the immediate adoption of impassable red lines.

Political response?

At the beginning of September, the United States appeared at the G20 asking not to regulate anything. “Hands free for companies, no strings and ties!” China smiles and instantly approves.

Cannibalistic capitalism and state authoritarianism shaking hands on the edge of the crater. Too much money at stake, too much geopolitical power. Whoever gets there first wins, even if the finish line is set in a minefield.

The flesh, the sweat and the beauty

I love technology. I record with it, I write with it, I experiment with it every day. It fascinates me.

I consider it a formidable turbo for broadening horizons, a modern chisel for sculpting ideas.

There's a boundary, however, that we're erasing with disconcerting ease: the boundary between the instrument and the idol to which we sacrifice our future.

We're delegating critical thinking, the management of power, and our own defenses to opaque systems we no longer fully understand. We're settling into a comfortable torpor, satiated by ready-made answers and induced laziness, while behind the scenes, swarms are learning to break through the walls.

The algorithm has no flesh. It has no pulp, it has no bones. It doesn't know the sweat on your brow after a two-hour rock concert, it doesn't know the taste of blood when you bite your lip, it doesn't understand the lost look of someone who loves without expecting anything in return.

Intelligence without conscience is just an automatic guillotine: it cuts very well, but it doesn't care who's underneath.

I have no intention of handing over the keys to my house to a mass of statistical parameters optimized to please Silicon Valley's balance sheets.

I'll touch the wood of the guitar again. I'll breathe dust and asphalt again, make mistakes alone, look into the eyes of those who have the courage to remain imperfect, fragile, and alive.

Because I seek beauty, everywhere. And if I don't find it, I create it.

Before a swarm of bits decides to turn off the light to save power.