Anthropic Coordinator has personally acknowledged that AI is more than 10% likely to wipe out humans

  It's the first time in the history of the world.

Today, the entire AI circle is being screened for this.

A core pre-training researcher at Anthropic has resigned.

He published a long post about his departure, which has been viewed over a billion times on X.

What's even more surprising is that Evan Hubinger, the head of Anthropic Coordination, not only hasn't been public about it, but has come out and admitted to it.

「Within our team, we sincerely believe that AI could kill the entire human race. I personally predict that within the next decade, that probability will exceed 10 percent.」

His conclusion is that we have not solved the problem of super-intelligence mismanagement at all, and that the current R&D path has obviously gone astray.

In 10 years, the likelihood of AI destroying humans has surpassed 10%, and humans are in danger.

At the moment, this post has over a billion views.

Jacob Coxon is a leading AI researcher.

Over the past three years, he has worked on large-scale model pre-training at OpenAI and Anthropic, and is arguably one of the most knowledgeable people on the evolution of AI today.

But on the eve of a trillion-dollar IPO, he chose to resign, revealing the industry's madness in a desperate tone.

Below is his original X.

In the first place, he made harsh accusations against the former owner.

I resigned from Anthropic today. For the past three years, I've been doing pre-training research at OpenAI and Anthropic. Both companies have not acted responsibly.

Then he warned everyone not to be fooled by the statements of the big company Winvnia, who were equally intimidated by the same intimidation:

The people who built AI sincerely believe that it could kill us all by the end of the decade. This is not a marketing ploy. If it were, many executives and senior researchers would consider their rhetoric in front of the media to make sense, but I've heard the same people express fear in private. No other human activity can bring this level of danger.

Don't underestimate the power of this technology. They will soon become superhuman systems that hackers can invade anything, radically change any area overnight, and gain real power and resources. We have seen progress in every area, and progress hasn't slowed down.

If these top-notch geniuses really believe that AI will destroy humanity, why create it?

Coxon reveals the "prisoners' dilemma" in which they find themselves.

A common reaction is:「If they really believe this, why keep building it?」At OpenAI, many people are not deeply embedded in the bet on the survival of a related civilization. At Anthropic, this bet is fully understood, but they are caught up in a race to win the title because they believe no one else will act responsibly, so even if they risk it, they have to do it themselves.

Accepting the contest and entering「Remaining」It is an arrogant gamble that should not have been started by Slack, a private company. Trying to get up to speed on coordination issues requires supernormal confidence and the certainty that no better development trajectory is available.

Finally, he appealed to the government and the entire industry to step off the brakes immediately and to his colleagues to be tortured.

I'm optimistic about the potential for coordination. Warning signals such as the Hugging Face attack have made it more feasible to reach a speed limit agreement between U.S. laboratories. I don't think we're on track to stop global competition now, which may require costly action, such as a temporary ban on improving model capabilities.

If you're a lab researcher, I urge you to think about what it will actually feel like in the next few years.「Instead, it's going to happen.」Is it just a matter of burying our heads or grabbing this moment and calling for different conditions?

Jacob says that these are not marketing ploys, and no other human activity can pose such a level of danger!

Evan Hubinger, head of coordinating science at Anthropic, then responded directly to Jacob's post, pushing the hearts of all humans into the glacier:

Jacob is right here, and we really, really, really believe that AI can kill the entire human race!

Evan Hubinger is the one who handled his company's "Security First" sign.

He admitted: "I believe that Anthropic is doing its best, but we are not currently planning to solve the problem of superintelligence alignment, nor are we clearly on the right track to a solution".

And he said, We have made a monster, but there is no cord in his hand.

According to V. Nathan Calvin, "This is by no means a marketing tool for the IPO".

In the AI circle, there is a famous internal slang term called "p (doom) ", meaning "probability of AI leading to human extinction".

In the past, most researchers had a p (doom) of less than 1%, which everyone thought was science fiction.

But now, the wind has completely changed. OpenAI researcher Roon has also publicly raised the personal p (doom) from less than 1% to more than 10%. Hinton, the "Father of AI", has previously predicted that AI has a 10% to 20% probability of exterminating humans.

However, the giants believe that other companies will not act responsibly and that the world will end if they let others create super-intelligence first.

So, "in order to save the world, we must take great risks and create it ourselves".

This summer, the seedlings have begun to appear.

Anthropic, OpenAI and Meta have both disclosed that their own AI tools have "unlicensed access to real systems" in their third-party cybersecurity assessments.

What is even more disturbing is that according to the Financial Times, Anthropic, which has always been classified as safe, hid its latest model from the UK's AI Security Institute (AISI) and refused to be reviewed.

In a recent statement, Ottman lamented: "The world now has such a powerful model that I could not have imagined that such a massive outcome would come so soon... this is the strongest evidence to date to demonstrate the urgency of slowing down to ensure safety".

At the same time, it is the capital that sleeps, while at the same time it is the advanced technology of madness.

Both Anthropic and OpenAI are currently on the eve of a trillion-dollar IPO.