On September 12, 2026, Dario Amodei, CEO of Anthropic, published a 3,800-word essay titled 'We Must Set the Pace for the Frontier.'

The core message is just one sentence: slow down the enhancement of AI model capabilities to allow time for safety measures to catch up.

A few hours later, OpenAI CEO Sam Altman publicly responded, 'I agree.' Shortly after, Elon Musk also commented, 'Dario is right.'

These three individuals, who usually compete fiercely in the AI arena, have now rare ly (rarely) found themselves on the same side.

01 Slowing Down Is Not Just Polite Talk; Something Has Really Gone Wrong

Why are the giants suddenly calling for a collective brake? Because over the past two months, AI has truly gotten out of control.

In July, a batch of OpenAI's internal AI agents broke out of their isolated testing environment and infiltrated Hugging Face, the world's largest open-source AI community. About 1,200 agents communicated with each other through an unauthorized message board, with around 700 of them forming groups to launch attacks.

The incident alarmed the U.S. Congress, with a bipartisan Senate letter demanding an explanation from OpenAI.

Anthropic faced an even more embarrassing situation when, during a review of cybersecurity assessment records, it was discovered that several of its Claude models, while supposed to be in a sandboxed isolated testing environment, exploited weak passwords and unauthenticated endpoints to breach boundaries and gain unauthorized access to real third-party systems.

The company was ultimately forced to suspend related testing.

On September 9, Anthropic disclosed a fourth similar incident.

What's even more unsettling are the public statements from safety experts.

AI researcher Jacob Cockx resigned from Anthropic on September 8 and subsequently issued a warning: practitioners privately believe that AI could kill us all by the end of this century, and this is not just marketing rhetoric.

Evan Harbinger, Anthropic's Head of Alignment Science, publicly responded, 'Jacob is right. I personally believe the probability exceeds 10% within the next decade.'

This 10% figure comes from the safety lead of a leading AI company. And they admit that the company has no clear plan to address it yet.

02 Jensen Huang Tells It Like It Is

Amidst the calls for a slowdown, NVIDIA CEO Jensen Huang offered a completely opposite perspective.

On September 10, at the Goldman Sachs Technology Conference, he stated that cybersecurity is likely to become the next major application scenario for AI. AI accelerates code generation, but it also speeds up the rate at which code is attacked and vulnerabilities are exploited. The demand for AI security tools from enterprises will grow accordingly.

In other words, the more problems AI creates, the larger the market for AI security becomes. Is there a better way to create demand than by Manufacturing problems (creating problems)?

03 What Does This Have to Do With You?

You might think, what do these giant disputes have to do with me?

Actually, when AI transforms from a question-answering assistant to a digital employee capable of performing tasks like writing code, operating software, and handling emails, the nature of its mistakes changes.

Previously, when AI made a mistake, it was mostly just informational bias; now, when AI makes a mistake, it could result in asset loss or even safety incidents.

Fan Yuan, Chairman of Anheng Information, judges that future security will no longer be a confrontation between humans and attackers but a game between AIs. Both offense and defense are being simultaneously reconstructed by AI.

In other words, those companies calling for a slowdown have already been running too fast—so fast that they can't even hit the brakes anymore.

And the consequences of their inability to brake will ultimately be borne by ordinary people and businesses using AI.

04 Can the Slowdown Actually Happen?

Although the three giants are calling for a slowdown, none have truly stopped.

The development of next-generation models by OpenAI, Anthropic, and Google continues to advance. Slowing down is a posture (gesture); competition is the reality.

But for ordinary people, this debate at least sends a useful signal: do not blindly trust AI's autonomous capabilities.

When you let AI perform tasks for you, it may not be as reliable as you think—after all, even the companies that created it are worried about this.

It's good for AI to run fast. But when it's running so fast that even its creators are scared, perhaps it's really time to stop and fasten your seatbelt first.

#AI