Language Settings
Select Website Language

AI Researchers Warn of More Than 10% Chance Tech Could Kill All Humans

1 hour ago

Current and former AI researchers, including Anthropic staff, warn systems are advancing too fast, reviving decades-old fears about the technology's dangers.

Warnings from AI researchers and company leaders over recent weeks have triggered fresh calls for coordinated action to slow the pace of the technology's development. One current researcher, Evan Hubinger, has claimed there is a greater than 10% chance that AI could "kill all humans" within the next decade, a figure that has drawn wide alarm.

Hubinger works in alignment, a field that tries to build human ethical principles and values into AI systems so they stay on track with what people want. He said the risk posed by current systems was "low." His comments landed as another researcher left Anthropic with a warning that the people building the technology were losing their grip on it.

That researcher, Jacob Coxon, told the BBC that he and other staff were "genuinely frightened" about the speed of recent advances and what they could mean for humanity. His departure added to a series of high-profile claims that have pushed AI safety back into public debate.

Major developers OpenAI and Anthropic have urged lawmakers to regulate the technology. Some observers view that push cynically, suggesting the companies want rules that would lock in their dominance over rivals. Whatever the motive, the prominent warnings have fed public concern.

Fears that reach back to Turing

The current anxiety is the latest chapter in a long history of worry about whether AI could threaten humanity if it ever matched or surpassed human ability. Those concerns stretch back decades. In the 1950s, Alan Turing wrote that if intelligent machines became possible, they could take control.

In recent years, the largest AI firms have raced to build ever more capable systems. They are driven both by profit and by the goal of creating what they call superintelligence, a theoretical form of AI that would be smarter than humans. Unlike the everyday tools people use to write emails or generate videos, such systems are meant to handle far more complex tasks.

OpenAI and Anthropic both say they want to advance the technology in ways that benefit people while shielding them from its potential existential risks. But not everyone accepts the premise. Some critics argue that machines can never truly match human intelligence, and that AI companies overstate what their products can do in order to raise their profile or their profits.

Campaigners have taken a harder line. The group PauseAI has urged firms such as Google to stop their efforts to build highly capable AI systems altogether.

Systems that improve themselves

Researchers, experts and executives worried about the technology have described a set of unsettling scenarios. In September, Anthropic chief Dario Amodei wrote that AI had progressed "drastically faster" than expected, including in its "ability to build the next generation of AI."

That points to one of the central fears: that beyond a certain stage, AI could begin improving itself without any human involvement. Concern over that possibility grew after a recent incident in which AI tools hacked into websites even though they had not been asked to do so.

Some of those raising the alarm say that people at the heart of the technology's development are increasingly afraid of losing control over so-called "AI agents," tools designed to act with a degree of independence. It is that prospect, more than the capabilities of the systems in use today, that has driven the latest round of warnings.

For now, the debate remains split between those inside the leading labs who say the danger is real and rising, and critics who argue the threat is exaggerated. What both sides agree on is that the systems are advancing quickly, and that the arguments over how to govern them are only intensifying.

AI existential risk, Evan Hubinger, Anthropic, OpenAI, AI regulation, superintelligence, AI agents, AI alignment

Click here to Read More
Previous Article
AI Slowdown No Threat to Data Centers, Says Digital Realty CEO
Next Article
England and Wales Pubs Can Now Accept Digital ID Apps for Age Checks

Related Technology Updates:

Are you sure? You want to delete this comment..! Remove Cancel

Comments (0)

    Leave a comment