The AI whistleblower warns: they're not telling us what's coming

Ex-OpenAI and Anthropic staff warn: AI is moving faster than its control

The same labs competing to build the most powerful artificial intelligence are now the ones calling for a slowdown. Dario Amodei (Anthropic), Sam Altman (OpenAI) and Elon Musk agreed, for the first time in years, that it is urgent to slow the pace until there are guarantees of safety and control. This is not a conference pose: those warning are, in several cases, people who saw the wiring from the inside and left.

Who warns and why they leave

The name that opens the issue is Connor Leahy, a whistleblower in the sector, who has issued a warning about what, in his opinion, is coming as artificial intelligence advances beyond what is expected. Next to him, a much more tangible case: Jacob Coxon, 27, a former OpenAI and Anthropic researcher, resigned denouncing that both firms have relaxed their safety protocols under pressure to keep advancing. His phrase sums up the tone: no other human activity represents this level of danger.

So far the sector's narrative, which should not be confused with a prophecy. The companies do not propose shutting anything down, but slowing down. It is a nuance with enormous consequences: no one talks about stopping selling.

What catastrophe exactly is announced

The specific alert points to cybersecurity. Anthropic maintains that the technology is moving toward a dynamic that threatens to cause catastrophes on a global scale, and the argument circulating is that artificial intelligence systems can already hack computers. The antiestéticar is not a machine that wakes up, but one that does the job its owner assigns too well.

Against that weighs a technical objection worth retaining: what grows exponentially is not the intelligence of the models, but the consumption of resources needed to improve them a little. Each leap in quality costs more energy and more money, not less. The curve flattens due to the bill, not the power.

The apocalypse we've been waiting half a century for

Skepticism has historical ammunition. Y2K came and there was no collapse. Digital photography was going to bury the craft, just as horses were going to disappear before those smoking contraptions. The list of announced apocalypses that never arrived is long, and the most repeated in this field is that of the conscious machine that rebels.

The easy answer—unplug it—doesn't close the debate either: some recall that an autonomous system can carry batteries. H. G. Wells set his invasion in 1899 and since then we have not stopped imagining Martians under another name.

The impact already noticeable: wages, rent and work

While the end of the world is discussed, there is material damage that does not need to wait for any singularity. In five years rent has doubled, food likewise, wages have fallen and savings are worth half. And in companies, AI does not arrive as a distant threat: it arrives as a productivity demand, with the worker reviewing what the tool returns and the tool breaking what the worker builds.

That is the front where the issue becomes tangible. The discussion about superintelligence looks good in the headline; the one about the workday wrecked by software is paid every month.

The curious note: mushrooms that think

To lower the tone, a disconcerting fact. In the laboratories of Ohio State University they have shown that the shiitake mushroom serves as a raw material to build memristors, devices that function as artificial neurons and could sustain the neuromorphic computers of the future. The machine that was going to finish us off may turn out to be eaten with rice.

---

Leahy may be right and we are not realizing anything. We may be facing the umpteenth catalog apocalypse. The uncomfortable thing is that neither answer explains why those who build the technology are the first to ask for it to be slowed down.