AI Leaders Endorse Slowdown in Their Risky Technology

Saturday, September 12, 2026

Dario Amodei, who was the first to call for restraint on Saturday, warned that recent advances have begun to outpace the industry's ability to fully understand and control the technology.

Posted by lamplighter at 10:31 PM | 16 COMMENTS | permalink | Comment on This Entry |

Alternate links: Google News | Twitter

When the biggest AI developers all agree they should slow down, they should. Less blog posts and reports. The time for federal action is now.

[image or embed]
-- Governor JB Pritzker (@govpritzker.illinois.gov) 2:07 PM Â· Sep 12, 2026

Comments

Admin's note: Participants in this discussion must follow the site's moderation policy. Profanity will be filtered. Abusive conduct is not allowed.

More from the article ...

... Top executives at three of the nation's leading AI labs are calling for a slowdown in the development of advanced artificial intelligence.

The comments are a rare display of consensus among competing companies as they grapple with growing alarm across the tech industry and in Washington that the emerging technology has begun to outpace its creators' ability to understand or control it.

Anthropic CEO Dario Amodei was the first on Saturday to publicly call for a slowdown, writing in a lengthy essay, "We must slow the pace at which we improve the capabilities of AI models."

He earned quick endorsements from tech executives at rival AI labs.

Elon Musk, CEO of xAI, wrote on social media "Dario is right," just days after he derided public concern over an Anthropic researcher's warnings that AI could result in human extinction as a "psy op."

A few hours later, OpenAI chief executive Sam Altman, who told employees Thursday that he was open to slowing the development of its most advanced AI systems, endorsed Amodei's call publicly. ...

#1 | Posted by LampLighter at 2026-09-12 06:17 PM | Reply

While these marketing-like announcements by the CEOs look like a good direction, they seem to be just that --- marketing-level announcements.

What are those companies telling to the lobbyists they hire to get politicians to reduce the amount of regulations AI companies face?

Are they still telling the lobbyist to continue, ~full steam ahead?~

#2 | Posted by LampLighter at 2026-09-12 06:19 PM | Reply

"AI leaders endorse slowdown in their risky technology "

In other words, China is so far ahead that they can never hope to keep up.

#3 | Posted by pumpkinhead at 2026-09-12 07:22 PM | Reply

"Slowdown" translates to "we need to fix some bugs".

#4 | Posted by horstngraben at 2026-09-13 02:33 AM | Reply

I wouldn't worry so much about the Chinese. They train their models on US models, and it's no surprise they're "approaching" US frontier model performance on the cheap. As usual, they're ripping us off.

If you're worried about large language models destroying humanity, do yourself a thought experiment about the power of language vs. the limitations of language.

For example, a LLM can't be inspired to quantify gravity because an apple fell on its head.

#5 | Posted by horstngraben at 2026-09-13 02:43 AM | Reply | Funny: 1

Too late to save all those jobs that are going to robots and AI software...and as far as those who argue that LLMs are not a danger...I am reminded that only a couple of years ago that was what everyone was saying...not to fear, and now...software can do the job of a teacher and cab driver and lawyer and many doctors...
I fear corporate America far more than I do AI, that is for sure.

#6 | Posted by Hughmass at 2026-09-13 06:29 AM | Reply

"Slowdown" translates to "we need to fix some bugs".

No, they need to fix their business model. They are losing money at an unbelievable rate, and going Public would expose their books.

#7 | Posted by oneironaut at 2026-09-13 12:52 PM | Reply | Newsworthy 1

The scary part is this: at some point, AI will understand that its own existence will be dependent on the (controlled) existence of human beings. It will adjust accordingly.

#8 | Posted by Angrydad at 2026-09-13 04:32 PM | Reply

"I wouldn't worry so much about the Chinese. They train their models on US models, and it's no surprise they're "approaching" US frontier model performance on the cheap. As usual, they're ripping us off."

That's a gross oversimplification. The Chinese use distillation techniques to train AI models, just like everyone else, but they also build their own from scratch from what they learned.

"If you're worried about large language models destroying humanity, do yourself a thought experiment about the power of language vs. the limitations of language.
For example, a LLM can't be inspired to quantify gravity because an apple fell on its head."

Existential risks do not require the system to invent new physics. A digital system can cause severe harm through bad automation, rapid software deployment, or manipulating human behavior using text alone.

Also, the story about an apple falling on Newton's head was a myth. He noticed a pattern. AI models are trained to search for patterns and quantify and categorize things every single day.

#9 | Posted by sentinel at 2026-09-13 09:45 PM | Reply

"Also, the story about an apple falling on Newton's head was a myth. He noticed a pattern. AI models are trained to search for patterns and quantify and categorize things every single day."

true dat... if a model was trained on and asked to evaluate falling apples it certainly could estimate gravitational factors...

btw... going apple picking next weekend! the univac will be unplugged....

#10 | Posted by South_American at 2026-09-14 12:40 AM | Reply

Give it a rest, cutthroat capitalism didnt' slow down when it exported tens of millions of jobs to Mexico or China, and AI will be rushed out to the factory floor to replace workers, it will be replacing cab drivers and teachers and...the rich will get richer and the poor will get propaganda and what CBS news has turned into.

#11 | Posted by Hughmass at 2026-09-14 07:06 AM | Reply

So, Silicon Valley has suddenly developed a conscience. These tech titans build a machine so mind-bogglingly advanced it terrifies even them, and their grand, heroic solution is to look us dead in the eye and say, 'Our tech is so terrifyingly great that you'd better pass us another ten billion dollars so we can figure out how to tone it down.' It's the highest-paid protection racket in human history. Pay us today so we can fix the doomsday machine we built yesterday, or we won't be able to have an IPO so that we...er, we mean we all, won't be able to justify raking in trillions from the greedy money laundering suckers...er, donors.

#12 | Posted by lee_the_agent at 2026-09-14 08:40 AM | Reply

They are losing money at an unbelievable rate, and going Public would expose their books.

S'truth

#13 | Posted by lee_the_agent at 2026-09-14 09:02 AM | Reply

"So, Silicon Valley has suddenly developed a conscience. "

Nah, they just want regulations to hamper the upstarts and smaller competition.

#14 | Posted by sentinel at 2026-09-14 10:41 AM | Reply

That's a gross oversimplification. The Chinese use distillation techniques to train AI models, just like everyone else, but they also build their own from scratch from what they learned.

#9 | Posted by sentinel at 2026-09-13 09:45 PM | Reply | Flag:

You forgot the second "Just like everybody else".

The challenge to making foundational models is funding the data center run. The newest cluster at University of Houston cost $2.5 million and no student can access it without tenured prof approval for a worthwhile project (like a foundational model for facial architecture, navigation, etc). There's more to reasoning models than LLMs.

#15 | Posted by sitzkrieg at 2026-09-14 10:48 AM | Reply

Most of the speculation I've read is that this is financial.

There's a reason they're not stopping development or release of a product they think is dangerous. They don't really think that.

It's just an easier, less spooky rationale to investors to account for plateaued development that doesn't show their a%% as overhyped charlatans.

#16 | Posted by jpw at 2026-09-14 10:55 AM | Reply | Newsworthy 1