Why are there concerns AI could threaten humanity, and how real are they?

ByLiv McMahon

Technology reporter

In caller weeks, warnings astir imaginable dangers of AI person spurred calls for coordinated enactment to dilatory the technology's development.

Claims by existent and erstwhile researchers that the tech volition endanger humanity - with 1 suggesting determination is simply a greater than 10% accidental of it "killing each humans" - person raised alarm bells.

Major developers OpenAI and Anthropic person urged lawmakers to modulate it, though immoderate possibly cynically suggest this determination aims to cement their dominance implicit rivals.

Regardless, the precocious illustration claims person caused concern. Here's what we know.

Some AI researchers accidental determination are worrying signs of systems processing successful their capabilities and powerfulness excessively quickly.

Evan Hubinger, made headlines with his content that determination is simply a greater than 10% accidental AI "could termination each humans" wrong the adjacent decade.

He works successful alignment, a tract which aims to physique quality ethical ideas and principles into AI. In different words, it aims to support it connected way with what humans value.

While helium said the hazard from existent systems was "low", his comments came arsenic different researcher discontinue Anthropic and warned those processing the tech were losing their grip connected it.

Jacob Coxon told the BBC helium and different unit were "genuinely frightened" astir the velocity of advancements and what it could mean for humanity.

These concerns are the latest successful a agelong enactment of fears astir AI's imaginable to endanger humanity should it turn to lucifer oregon surpass quality abilities.

Such worries day backmost decades, with Alan Turing penning successful the 1950s that should intelligent machines go possible, they could instrumentality control.

In the past fewer years, large AI companies person been racing to make progressively susceptible systems, some for nett and successful pursuit of creating what they telephone ace quality - theoretical tech which is smarter than humans.

Unlike the systems and tools galore radical usage regular to implicit tasks similar penning emails oregon generating videos, ace intelligent AI systems should beryllium capable to grip acold much analyzable tasks.

OpenAI and Anthropic accidental they purpose to beforehand specified tech successful a mode which benefits humans and support them from its existential risks.

But immoderate critics judge machines tin ne'er genuinely lucifer humans. They accidental AI firms exaggerate the tech's abilities to boost their prominence oregon profit.

AI researchers, experts and bosses disquieted astir the technology's imaginable to endanger humanity person outlined unsettling scenarios.

Dario Amodei wrote successful September, external it had precocious "drastically faster" than expected, including successful its "ability to physique the adjacent procreation of AI".

This is the interest that AI, aft it has reached a definite point, could theoretically statesman improving itself without quality involvement.

Such fears were sparked aft a caller incidental saw AI tools hack into websites, contempt not being asked to.

Some acrophobic researchers person said those astatine the bosom of its improvement are progressively fearful astir losing power of systems known arsenic "AI agents".

These are tools fixed the quality to execute tasks and actions independently.

Other scenarios person included highly intelligent AI systems being utilized to amplify oregon pb espionage, cyber-attacks oregon biologic warfare, oregon disrupt economies - whether by mishap oregon design.

However, specified warnings are prophetic, hypothetical and, for some, complete hyperbole, external.

Some experts person besides criticised focusing connected existential risks alternatively of those already playing retired - specified arsenic representation procreation tools being to utilized make non-consensual nude deepfakes of women, misinform radical and amplify scams.

A cross-party radical of MPs and peers has published a study outlining quality rights risks posed by AI, saying caller laws are needed to "address the standard and seriousness" of specified threats.

Dario Amodei of Anthropic and Sam Altman, main enforcement of OpenAI, person some called for regularisation and a "slowdown" of the tech's development.

It is not the archetypal clip bosses oregon wider manufacture voices person clamoured for specified action.

But questions person continued to equine implicit what immoderate benignant of "slowdown" - besides backed by SpaceX brag Elon Musk - would really look like.

Amodei said it would not mean "halting exemplary grooming oregon method progress".

Instead, helium wants firms to "take capable time" to safeguard the AI they are developing, and to inquire for "third enactment evaluators" to cheque their work.

In a post connected X, external, Altman said determination was a request to gait improvement but added "when we speech astir 'pacing' we bash not mean 'stopping'."

"Progress has been accelerated and volition proceed to be," helium said.

"But it should beryllium slower than it different could beryllium - interventions similar information cases and monitoring person important costs."

The OpenAI co-founder besides said companies did not - and would not - hold for governments to act.

President Donald Trump has antecedently lauded the benefits of AI and championed US companies starring connected its development.

His medication has signed enforcement orders to recognise the tech's "strategic importance" to the system and accelerate its development.

"We judge we're successful an AI race, and we privation the United States to triumph that race," the US administration's crypto caput David Sacks told reporters successful July 2025.

But successful the look of caller warnings astir its imaginable interaction connected humanity, President Trump downplayed risks posed by AI.

"You person a batch of precise antagonistic forces that are bringing it up that shouldn't beryllium bringing it up and they're bringing up things that won't happen," Trump said during a sojourn to Ireland.

He did not straight code the thought of a slowdown, but told reporters: "We're starring China connected AI... and, frankly, I privation to support it that mode due to the fact that whoever wins AI, wins."

China had kept quiescent successful effect to the question of dire warnings from researchers astatine US companies.

In the look of fierce contention and limitations connected its entree to US components and chips being utilized to physique AI infrastructure, the state has spurred homegrown improvement of the technology.

At the aforesaid time, it has called for greater planetary governance of AI.

The rotation retired of capable, open-source AI systems and tools by Chinese labs has shaken Western markets and developers successful caller years.

Amodei warned successful calling for a slowdown of AI improvement that China's advancement and proficiency successful AI would make nationalist information risks for the US.

The Chinese overseas ministry criticised the comments, calling them narratives of threat, confrontation, and malicious contention that served nary one's interest.