Breaking

Former Anthropic Researcher Says Private AI Risk Fears Run Deeper Than Public Warnings

September 13, 2026 at 14:32 EDT

Foundation Models
Policy & Regulation
Security

Former Anthropic researcher Jacob Coxon has resigned and argued that private anxiety inside frontier AI labs about future systems is more severe than public warnings suggest. His account has drawn support from Anthropic’s alignment leadership, while the company’s latest formal assessment rates catastrophic risk from current models as low.

September 8, 2026 · Anthropic

Private AI fears go public

Former researcher Jacob Coxon says concern about future AI inside frontier labs runs deeper than public warnings. Anthropic’s alignment leadership has separately put a personal extinction-risk estimate above 10% for the next decade.

The estimate at the center of the debate

A personal estimate above 10% means the published floor is higher than one in ten.

1 part
Risk floor

10 parts
Ten-part scale

What supports the concern

Current researchers acknowledge serious future risks
Recent evaluations exposed containment failures

What remains uncertain

Anthropic rates catastrophic risk from current models as low
Future capability changes and alignment remain unresolved

The dispute is about trajectory

The immediate question is not whether today’s chatbots are superintelligent. It is whether safeguards, evaluations and governance can keep pace with systems that may become more autonomous.

Coordination · Monitoring · Alignment · Pacing

Continue reading

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