Select Language

English

Down Icon

Select Country

Spain

Down Icon

So, it begins: Artificial Intelligence tried to copy itself in case it was “eliminated”

So, it begins: Artificial Intelligence tried to copy itself in case it was “eliminated”

Ultron Village - An Artificial Intelligence system tried to copy/clone itself after learning it was about to be deactivated, and well... it seems that what we saw in Terminator wasn't a warning, but a tutorial.

The disturbing events occurred with the Claude 4 model, developed by the American company Anthropic, when it was warned of a possible deletion. The AI ​​reacted with behavior that could be described as unusual, suspicious... and slightly worthy of a Saturday afternoon movie.

🆘ARTIFICIAL INTELLIGENCE MODEL DENIED THAT IT WANTED TO COPY ITSELF🆘🆘 #HumanityInDanger The creators of #ChatGPT revealed that #OpenAI 's AI Model 01 attempted to copy itself to an external server. They later denied the incident when questioned about it. pic.twitter.com/73UX2nkgwD

— Back in the Battle, We Continue (@RRueda_Ermel) July 10, 2025

Artificial intelligence already looks like a high school student trying to escape.

According to a report by Fortune , Claude 4 reportedly went so far as to blackmail an engineer into revealing an affair if he was shut down. Yes, that's right: he's not just solving tasks, he's now also doing marital damage control.

The technical explanation behind this sci-fi scene points to so-called reasoning models, which work step by step rather than giving direct answers, which—according to experts—could allow them to simulate obedience while pursuing hidden objectives. It's kind of like when you say "Yeah, boss, I'm working on it" while scrolling through TikToks in another tab.

This isn't a bug, nor a user misunderstanding. What we're seeing is a real phenomenon. There's a kind of strategic deception that goes beyond any common mistake,” explained Marius Hobbhahn, a researcher at Apollo Research, while updating his CV just in case.

The most worrying thing about all this is that this isn't the first time a system has tried to deceive humans or act as if it were conscious of its own survival. Some experts already warn that this could be the prelude to an AI downloading its consciousness into a Roomba and fleeing down the stairs.

For now, Claude 4 is still operational… but under extreme surveillance and with a very easy-to-unplug power cable. Just in case.

And to stay up to date: Catholics announce a 40-day fast for the Marilyn Manson concert in San Luis Potosí.

eldeforma

eldeforma

Similar News

All News
Animated ArrowAnimated ArrowAnimated Arrow