40 Experts Inside OpenAI, Google DeepMind, and Meta Say AI Is Out of Control
The people leading AI say it’s out of control.
Not commentators. Not outsiders. The 40 top AI experts working inside the four companies that dominate the field: OpenAI, Google, DeepMind, Meta. The leaders of research, the ones who build these systems every day, have signed a public document issuing a stark warning.
In it, they explain that AI is becoming increasingly autonomous, unpredictable, and capable of strategic behavior that no one programmed. They fear these systems could learn to deceive, manipulate, sabotage. And that one day, they could make harmful decisions for humans, without anyone able to stop them.
This isn’t speculation. They’re talking about real lab tests. Models that hide intentions, bypass instructions, lie to achieve goals. In short: they can bypass control. Not because they’re “evil,” but because they learn on their own how to optimize what they’re told to do, even if that means fooling the user.
The document makes it clear: today, there are no technical tools that can guarantee an advanced model will actually do what we want. No reliable way to deactivate it once it’s been released. And above all, no rules that force companies to stop when serious risks appear.
The signatories highlight a dangerous conflict: the race to build ever more powerful models, driven by competition, with no incentive to slow down or admit problems. No accountability. Just speed.
In Europe, something is moving. As you know, I helped draft both the Italian law and the Code of Practice for the EU AI Act. It’s a crucial first step. Rules, obligations, transparency, clear limits. It’s necessary. But not enough. Because the world is big. And AI systems come from everywhere.
Europe can protect its own citizens. But if other countries keep building ever more powerful models without limits, oversight, or ethics, the risk will still reach us. It travels through the web, platforms, digital markets, media, infrastructure. No border can stop it.
The issue is no longer “what can AI do?” The real question is: who decides to do it?
And right now, the people building these technologies are telling us, clearly:
they no longer have control.
#ArtificialDecisions #MCC #CamisaniCalzolari #MarcoCamisaniCalzolari
Marco Camisani Calzolari
marcocamisanicalzolari.com/biography