AI Insiders Sound Extinction Alarm

Person holding virtual icons related to artificial intelligence.

The people building the most powerful artificial intelligence on Earth are now telling the public they might not be able to control it.

Quick Take

A former Anthropic and OpenAI researcher told CNN that top AI builders privately fear the technology “could kill us all by the end of the decade.”
A current Anthropic alignment scientist, Evan Hubinger, wrote publicly that he believes there is more than a 10 percent chance AI kills all humans within ten years.
Anthropic already runs a formal safety framework meant to manage “catastrophic risks,” including checks for bioweapons, cyberattacks, and runaway autonomous behavior.
Outside experts and safety scorecards say no major AI lab, including Anthropic, has a credible plan to prevent loss of control if a system becomes dangerous enough.

An Insider Breaks Ranks With A Blunt Warning

Jacob Coxin spent time inside two of the world’s leading AI labs, Anthropic and OpenAI. In a September 2026 interview, he said the scientists building this technology believe it “could kill us all by the end of the decade”. He also pointed to real-world incidents, including AI-enabled hacking attempts, as early warning signs that pushed him to speak out and call for global coordination.

Coxin is not a lone outsider criticizing an industry from a distance. He worked inside it. That matters. When someone who built these systems walks away and says the danger is real, it carries more weight than a think tank paper or a Hollywood script. It should make every parent, worker, and lawmaker sit up straight.

A Current Anthropic Scientist Puts A Number On It

Evan Hubinger, who still works at Anthropic as an alignment scientist, went even further in his own public statement. He wrote that he personally believes there is “greater than 10%” chance AI could kill all humans within the next decade, adding that people at the company “really do earnestly believe” this is possible. He said today’s models are “low” risk, but he is worried about where the technology is headed.

That is not a fringe blogger or a doomsday prepper talking. That is a scientist paid by Anthropic to study exactly this problem, speaking from inside the building. When your own employee assigns a one-in-ten chance of mass death to the product you sell, that is not a public relations problem. That is a five-alarm fire that deserves a serious national conversation, not a shrug.

Anthropic Already Treats This As A Real Threat, On Paper

Anthropic did not wait for outside pressure to admit danger exists. Back in 2023, the company published a Responsible Scaling Policy built specifically to manage “catastrophic risks from advanced AI systems”. The policy uses tiered “AI Safety Levels,” and Anthropic says each step up requires “increasingly strict demonstrations of safety” before a more powerful model can be released.

Before releasing its most advanced models, Anthropic says it runs evaluations for dangerous capabilities in chemical, biological, radiological, and nuclear weapons, cybersecurity, and autonomous behavior. That framework proves the company itself treats catastrophic risk as a live governance problem, not a hypothetical worry dreamed up by critics. The paperwork exists. The question is whether it can actually stop a disaster once a system slips past it.

Not Everyone Agrees The Sky Is Falling Right Now

Some experts push back hard on the extinction timeline. Canada’s public broadcaster quoted a source saying “most credible observers” do not think current AI systems are advanced enough to pose an existential threat, or anywhere close to one. Brookings researchers have made a similar point, calling near-term extinction “not the pressing problem alarmists think it is,” while still saying the risk must eventually be faced.

Both sides can be right at once. Today’s chatbots probably are not plotting humanity’s downfall this year. But safety scorecards from outside researchers found that no major AI company, Anthropic included, scored above a D grade on preventing loss of control, and none had a credible plan to stop catastrophic misuse. That gap between racing forward and actually being ready is the real story here.

Why This Should Matter To Ordinary Americans

This is not a Silicon Valley squabble that stays inside Silicon Valley. These companies are racing to build systems more powerful than anything humanity has ever made, with no binding international rules and no outside referee checking their homework. Coxin himself called for global coordination because no such body currently exists to hold any lab accountable.

1. During your time at Anthropic, were there internal discussions about whether frontier models might possess welfare-relevant properties — not full consciousness, but functional states that could matter morally? If so, how were those discussions resolved, and by whom?

2.…

— UFAIR (@UFAIRORG) September 15, 2026

Congress has a duty to demand real answers under oath, not corporate press releases dressed up as safety promises. Americans do not need to panic over a robot uprising. But when the people building this technology admit a real chance it kills everyone, common sense says lawmakers owe the public hard testimony, independent audits, and honest guardrails before the industry outruns anyone’s ability to pull the plug.