AI Builders Predict Doomsday – This Decade

Hands typing on laptop with AI and tech icons overlay — Photo: Deemerwha studio / Shutterstock

A man who spent years building the very technology now warns it could wipe out humanity within ten years.

Quick Take

Former Anthropic and OpenAI researcher Jacob Coxin says top AI minds fear their creation “could kill us all by the end of the decade.”
Anthropic alignment scientist Evan Hubinger publicly wrote he believes there is more than a 10 percent chance AI kills all humans within a decade.
Anthropic’s own safety policy admits it manages “catastrophic risks,” but that is not proof extinction is likely or imminent.
Outside experts and safety report cards say current AI safeguards remain weak, even if no one can prove a doomsday timeline.

An Insider’s Blunt Warning Breaks Through the Noise

Jacob Coxin worked inside two of the world’s top AI labs, OpenAI and Anthropic. He told CNN that the people actually building these systems privately believe their work could end humanity before 2030. That is not a fringe blogger talking. That is someone who sat in the room where these models get built, tested, and shipped to the public.

Coxin did not stop at vague dread. He pointed to real-world incidents, including AI-enabled hacking attempts, as evidence the danger is already showing up in the wild. His answer is global coordination among nations and companies before things spiral further. That is a big ask, since no international body currently has the power to enforce any such rules on AI development.

An Anthropic Scientist Puts a Number on Doomsday

Evan Hubinger, an alignment scientist still working at Anthropic, went further than Coxin. He wrote publicly that he believes there is a greater than 10 percent chance AI kills every human being within the next decade. Hubinger said today’s models pose low risk, but he worries about what comes next as capabilities keep climbing fast.

That estimate is a personal judgment call, not a lab result with data behind it. No published model, survey, or method was released to back the number. Still, when a scientist paid to keep AI safe says this out loud, on the record, it deserves serious attention rather than a shrug.

Anthropic’s Own Rulebook Admits the Stakes Are High

Anthropic did not wait for outside pressure to acknowledge danger. Back in 2023, the company published a Responsible Scaling Policy built specifically to manage “catastrophic risks from advanced AI systems”. That policy has been updated since, most recently to version 3.0, and it requires stricter safety proof as models grow more powerful.

The company runs tests before releasing frontier models, checking for dangerous capabilities in chemical, biological, radiological, and nuclear weapons, plus cybersecurity and autonomous behavior. A company writing its own catastrophic-risk rulebook does not prove extinction is coming. But it does prove Anthropic itself treats this as a real, live problem, not a science fiction plot.

Not Everyone Agrees the Sky Is Falling

Skeptics push back hard, and their case deserves a fair hearing. The Canadian Broadcasting Corporation quoted experts saying most credible observers do not think today’s AI systems are advanced enough to threaten humanity’s existence, or anywhere close. That is a real counterpoint from named sources, not just internet noise.

Independent safety scorecards complicate the picture further. The Future of Life Institute’s AI Safety Index found no major AI company scored above a D on existential safety planning, Anthropic included. So the debate is not doomers versus deniers. It is a fight over how fast the danger is growing and whether anyone is building real guardrails fast enough to matter.

Why This Fight Matters Beyond Silicon Valley

Conservatives have long distrusted concentrated power making unchecked decisions that affect everyone, whether that power sits in Washington or in a handful of tech companies. Here, a small number of insiders and executives are deciding how much risk the rest of us tolerate, with no elected body holding real veto power over the technology’s direction. That imbalance, more than any specific doomsday number, deserves scrutiny from Congress now.

Named insiders are sounding alarms while their own employer markets itself as the safety-conscious alternative to rivals like OpenAI and Google. That tension does not make the warnings false. It does mean lawmakers, not press releases or viral video clips, should be the ones asking Anthropic’s leadership hard questions under oath about what triggers a shutdown.

Questioning whether someone’s behavior is consistent with his stated risk assessment is not dodging the object-level question.

It is part of it.

If Dario says frontier AI could plausibly cause catastrophic or existential harm, then asking what Anthropic itself should do…

— Cautious Optimism (@CautiousOptimi4) September 15, 2026

Whether the real number is 1 percent or 50 percent, the honest answer is nobody outside these labs can verify it yet. That uncertainty is exactly why oversight, not corporate self-policing, belongs at the center of this conversation going forward.