Breaking

Anthropic Alignment Lead Puts AI Extinction Odds Above 10% This Decade as IPO Nears

September 13, 2026 at 09:33 EDT

Foundation Models
Funding & M&A
Policy & Regulation

Anthropic’s alignment-science lead said he personally estimates a greater than 10 percent chance that artificial intelligence “could kill all humans” within the next decade, a warning that has intensified a fight over how fast frontier systems should be built as the Claude maker prepares to go public.

September 2026 · Anthropic

A 10% extinction warning, mid-IPO

Anthropic’s alignment-science lead says he personally sees more than a 10% chance AI could kill all humans within a decade, as the lab prepares to go public without a plan to align superintelligence.

Same unit, drawn to scale

Percent fatality-style odds. Column height equals the number.

Surgery fatality bar some investors say already means no

>10%

Hubinger personal odds AI kills all humans this decade

Takeaway: the personal figure is at least twice a 5% fatality threshold some market commentators treat as already unacceptable.

>10%

Personal decade extinction odds

Low

Risk from present models

$965B

Last private valuation

Mid-Oct

Expected IPO marketing window

How the warning landed

Coxon resigns

Accuses labs of racing to self-improving superintelligence

→

Hubinger agrees

Puts personal odds above 10% this decade

→

No SI plan

Company not clearly on track to align superintelligence

→

IPO fight

Investors and officials argue over pace and listing

Inside the labs

Fear described as earnest, not a stunt

Senior staff said to be more concerned. Hinton called 10% not unreasonable. Coxon says the next year or two is crunch time and wants coordinated limits on recursive self-improvement.

Pushback

Critics see a doom loop — or a pitch

A Pentagon official dismissed the warning. A UN AI adviser said some of it could be marketing around listings. Gary Marcus said he saw no realistic path to killing all humans.

Key admission

“We do not yet have a plan to solve alignment for superintelligence.”

Present models
Risk rated low in the latest risk report

The worry
Superintelligence from recursive self-improvement

Company line
Benefits and unprecedented risks; wants verifiable pacing

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.