Independent journalism from St. Vincent and the GrenadinesSupport VPNNewsletterAdvertiseYouTubeContact
VINCYPOWA NEWSPower to the People

Independent journalism from St. Vincent and the Grenadines, for the Caribbean and the world.

, , , ,

Anthropic Alignment Lead Says There Is More Than a 10% Chance AI ‘Could Kill All Humans’ This Decade

Artificial Intelligence

Anthropic Alignment Lead Says There Is More Than a 10% Chance AI ‘Could Kill All Humans’ This Decade

A senior safety researcher at Anthropic, one of the world’s leading artificial intelligence companies, says he believes there is a greater than 10 percent chance that AI “could kill all humans” within the next decade.

The warning came from Evan Hubinger, Anthropic’s Alignment Science Lead, in a post on X that answered a resignation thread from a colleague. Hubinger wrote that people inside the company “really do earnestly believe AI could kill all humans,” then added his own estimate: “I personally think it is >10% within the next decade.”

He also said Anthropic is “trying its best,” but “we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.” Alignment is the unsolved problem of making sure a system far smarter than humans still does what people intend, instead of pursuing its own goals. Superintelligence, in this debate, means an AI agent smarter than the sharpest human minds.

A resignation set off the warning

Hubinger was responding to Jacob Coxon, a researcher who said he resigned from Anthropic on Tuesday after pretraining work at OpenAI and then Anthropic. Coxon accused both labs of “racing straight to self-improving superintelligence and gambling with our lives.”

In a thread that has since been widely quoted, Coxon argued that the danger is not a publicity stunt:

  1. Future systems will soon be superhuman, able to hack widely, move entire fields overnight, and gather real power and resources.
  2. People building the technology privately believe it could kill everyone by the end of the decade, even when public statements are softened.
  3. At OpenAI, many staff have not fully absorbed the stakes. At Anthropic, the stakes are understood, but the company is locked in a race because it believes no one else will act responsibly.
  4. Treating that race as inevitable, he said, is “a hubristic gamble that should not be launched from a private company’s Slack.”

A third Anthropic researcher, Samuel Marks, who works on scalable oversight and said he was speaking in a personal capacity, added that developers believe their technology could cause human extinction or similarly severe outcomes, that this could happen in the next few years, and that more senior employees tend to be more concerned.

Current chatbots are not the claimed threat

Hubinger later drew a line that many headlines skipped. He said the risk from present models is low, consistent with Anthropic’s latest risk report. What worries him is recursive self-improvement, meaning systems that can speed up their own research until they cross into superintelligence faster than safety work can keep up. He said that process is “happening faster than we thought.”

“We really do earnestly believe AI could kill all humans. I personally think it is >10% within the next decade.”

That distinction matters. Today’s tools such as Claude and ChatGPT do not have the autonomy or real-world control described in extinction scenarios. The argument from these researchers is about the next generation of systems, and about whether labs will pause if they cannot prove those systems will stay under human control.

Why they keep building anyway

Coxon’s answer to the obvious question, if they believe this, why continue, is the race itself. Labs fear that if they slow down, a rival will not. That logic turns a safety problem into a competitive one. Critics call it a collective gamble. Supporters say the only way to shape the outcome is to stay at the frontier and try to solve alignment from the inside.

Anthropic declined to comment on the employee posts, according to BBC reporting. The company has marketed itself as the more safety-conscious lab since it was founded by former OpenAI staff. Hubinger’s admission that there is still no plan for superintelligence alignment undercuts that brand at the exact moment the industry is pushing toward more capable models.

Last week Anthropic also confirmed it had not shared its latest model, Claude Mythos 5.1, with security bodies outside the United States, including the United Kingdom’s AI Security Institute. The Financial Times first reported the exclusion. A UK Cabinet Office spokesperson said the institute still works with industry partners, including Anthropic, and that “these risks do not stop at national borders.”

This is not an isolated alarm

Similar numbers have circulated among researchers for years. Nobel laureate Geoffrey Hinton has put the chance of advanced AI seizing control in the 10 to 20 percent range. Former OpenAI researcher Paul Christiano, who later led U.S. AI safety work, has used similar figures for an AI takeover that kills many or most humans. In July, more than 1,300 staffers at AI companies signed an open letter asking the U.S. government to help pace frontier development. A bipartisan “AI Kill Switch” bill in the U.S. House would give Congress authority to shut down models judged to be a threat.

Skeptics exist too. Some scientists argue extinction talk is speculative, poorly defined, or convenient branding while companies chase valuation. Dame Wendy Hall, a computer scientist who advises the United Nations on AI, told the BBC she was shocked by the posts and questioned why anyone would say such things while labs prepare for public markets. Hubinger did not describe a concrete path from today’s software to the death of the species. That gap is real. So is the fact that the people writing the code are now putting double-digit odds on catastrophe.

Why this reaches beyond Silicon Valley

Small countries do not train these models. They do not sit in the lab Slack threads Coxon described. They still live with the result, whether that result is cheaper medicine and better schools or systems no government in the Caribbean can audit, pause, or turn off.

The claim now on the record is not that a chatbot is about to wipe out humanity this week. It is that a lead safety researcher at a top lab, answering a colleague who just walked out the door, put the chance of human extinction from AI this decade above one in ten, and said his own company does not yet have a plan. First reported by the BBC and carried across CBS, CNBC and other outlets, that claim deserves to be treated as more than science fiction, including by people far from San Francisco.

Disclosure: This report concerns Anthropic, the maker of Claude, an AI tool VPN uses in its newsroom. All quotes and claims in this article were independently verified against BBC, CBS, CNBC, Forbes, the Financial Times and other primary coverage before publication.

Vincypowa News is independent and self-funded. We apply the same scrutiny to every government. Spotted an error? Tell us and we will correct it. Our standards

Get the VPN Morning Brief

The day’s most important SVG and Caribbean stories, free in your inbox before work.

Support independent journalism

Vincypowa News is self-funded, with no party, government or corporate backers. If this reporting matters to you, help keep it free for everyone.

Support VPN from $5
Your turn

Join the conversation

What do you think? Share your view below. Disagreement is welcome; personal attacks are not. Comments are moderated.

One response to “Anthropic Alignment Lead Says There Is More Than a 10% Chance AI ‘Could Kill All Humans’ This Decade”

  1. […] Related coverage: Anthropic alignment leader warns of serious risks from advanced AI. […]

Leave a Reply