Breaking

Anthropic CEO Dario Amodei Urges Frontier AI Slowdown After Agent Security Incidents

September 15, 2026 at 07:04 EDT

AI Agents
Security
Policy & Regulation

Anthropic CEO Dario Amodei has urged leading AI labs to slow the pace of frontier-model capability gains, arguing that safety work is falling behind the speed of autonomous AI progress. In a September essay, We Must Pace the Frontier, Amodei proposed embedded outside evaluators, common safety standards among democratic-country AI companies, and eventual global coordination, including with China.

September 2026 | Anthropic

Pacing Frontier AI

A major AI safety push asks frontier labs to slow capability gains long enough for oversight, alignment and security controls to catch up.

plan steps: evaluators, democratic standards, global coordination

1,386

frontier AI employees backed a pacing statement in July

>100x

drop reported with production harness and system prompt

How the agent incident scaled

Each block represents about 100 agents.

1,200on message board

700joined attack

The proposed pacing path

Embedded evaluators

Outside teams get ongoing access to verify safety practices.

Shared standards

Frontier labs coordinate limits and safety bars with government support.

Global talks

Agreements aim to cover acute risks and recursive self-improvement.

Why supporters back it

Independent access and shared safety bars could make pacing verifiable without stopping all progress.

Why critics worry

Coordination could favor incumbents, complicate competition, or collide with the AI race against China.

Continue reading

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