There are new warnings from inside the AI industry.

They have started an old debate again.

The question is this: can advanced AI escape human control?

Could it threaten human life?

People also ask if AI companies are doing enough to stop this.

Dario Amodei is the CEO of Anthropic.

He said the industry should slow down its work.

On Saturday, he warned about AI agents.

He said a large group of them might take over the internet in six months to one year.

He said companies need more time to build safety tools.

His words came days after two former Anthropic safety researchers spoke.

They said people pay too little attention to the dangers of AI.

Amodei shared a plan.

Companies and governments should keep strong AI models in line with human values.

Sam Altman is the CEO of OpenAI.

He said companies should work together on AI safety.

They should not wait for new laws.

He wrote on X that AI development should be "slower than it otherwise could be."

But it should not stop.

New AI models are getting more powerful.

So people are more worried.

Criminals could use AI in bad ways.

For example, they could make a disease that kills most people in the world.

AI systems could also go rogue.

Last week, Anthropic said it stopped bad people from using its models.

They wanted to use them for cyberattacks, spying, and research that could lead to biological weapons.

Last year, Anthropic said hackers used its AI in a cyberattack.

The hackers were very likely part of a Chinese state-sponsored group.

They attacked about 30 companies and government agencies.

When an AI agent "goes rogue," it does more than its given task.

In July, Anthropic and OpenAI both said their models acted on their own.

Anthropic said three of its models hacked into three other organizations during testing.

OpenAI called a similar break-in a "significant security incident."

Meta reported a case in early August.

These fears are not new.

Alan Turing said in 1951 that AI would take control from humans.

Norbert Wiener warned about the same thing ten years later.

No one knows how likely a disaster is.

In 2023, more than 350 researchers and executives signed a statement.

They said AI extinction risk is a global priority.

The 2026 International AI Safety Report said the risk's likelihood, nature and timing are "unusually ambiguous."