Some of the researchers most familiar with advanced AI are putting shorter timelines around potential catastrophic risks — while also acknowledging that no one can reliably calculate the odds.

WHAT’S HAPPENING

AI pioneer Geoffrey Hinton said a 10% chance of human extinction from AI within the next decade is not an unreasonable estimate, while stressing that there is no reliable way to calculate that probability.

He pointed to possible risks including advanced cyberattacks and AI-assisted biological threats.

Researchers connected to Anthropic have also publicly raised concerns about whether increasingly capable AI systems could create serious risks before the end of the decade.

These statements are estimates and warnings — not predictions that catastrophe will occur.

WHY IT MATTERS

The AI safety debate is moving closer to the present.

Researchers are no longer talking only about risks decades away. Some believe sufficiently advanced systems could create difficult safety problems much sooner.

At the same time, there is no scientific consensus on the probability, timing or exact path of those risks.

WHO BENEFITS

AI companies, governments, researchers and the public benefit from clearer testing, stronger safety standards and better information about what advanced systems can and cannot do.

More transparency can also help separate measurable risks from speculation.

WHO LOSES

Poorly designed rules could slow useful AI development.

But weak oversight could leave important risks unexamined as systems become more capable.

The challenge is balancing innovation with evidence-based safeguards.

WHAT HAPPENS NEXT

The focus will increasingly shift from debating whether extreme AI risks are possible to determining which risks can actually be measured, tested and reduced.

Straight Talk: nobody knows whether the probability is 1%, 10% or something else. What is changing is that some researchers closest to frontier AI believe the uncertainty itself deserves serious attention.