Sam Altman upset by GPT – 5: “There is no one left to stop this”

Sam Altman, CEO of Openai, recently said that a test moment with the GPT -4 successor, which can undoubtedly qualify as GPT – 5brought it back to a historical image: that of the physicists of the Manhattan project, realizing a little late what they had set in motion. “There was a moment that really marked me. It reminded me of the Manhattan project”he explains. It is not a shock formula to make the buzz. It is an admission, that of a deep discomfort in the face of a technology that has become too fast for humans.

GPT – 5: a model that goes “very quickly”

What strikes Altman is not only the raw power of the model. It is the speed of progression. “There is something in this system that makes it really very fast. It makes me nervous” he says. And he doesn’t say that lightly. It is a visceral reaction, that of a builder who looks at his creation and wonders if she is not already escaping him. He continues: “I think it is healthy to be a little afraid of the technology that is built.”

GPT – 5 has not yet been publicly deployed, But its launch is scheduled for early August. The internal test phases, almost completed, suggest a much more agile, reactive model, perhaps even strategically autonomous on certain tasks. An AI that even “surprises” its own designers.

“There is no one left to stop this”: the AI advances without pilot

Altman is not only worried about the tool. What concerns him is the void around. “There is no one left to stop this.”
This terrible sentence says a lot: the development of artificial intelligence advances faster than its regulation. There are no international control organization, no interoperable technical safeguard, or consensus on the limits to be installed. It’s a race. And in this race, even the leaders raise their arms to say: we no longer know where we are going.

Some will say that he exaggerates, that he dramatizes. But you have to replace your remarks: Altman is not for a bioethics committee. He does not seek to alert regulators. He speaks frankly, almost with broken sticks. And what he shares is a reality on the ground: current models learn quickly, improvise, and offer results that sometimes escape explicit logic.

This does not mean that GPT – 5 is dangerous. But that means that its poorly supervised deployment could be. We have seen what GPT – 4 could already produce – convincing hallucinationshuman style clones, malicious code generators. It is therefore reasonable to wonder about what the next version will make accessible on a large scale.

“It’s a good thing to be a little afraid”repeats Altman. This lucidity, which we would like more frequent in the sector, could precisely be the start of a start. Recognizing that these models are no longer only tools, but complex systems with sometimes emerging behaviors, this is the starting point for an ethical, political, legal debate.

Altman, hollow, asks us a question: is it still time to think, or are we already condemned to passively support a technology that we no longer govern?

Want to save even more? Discover Our promo codes Selected for you.