Breaking

Anthropic CEO Urges Slower Frontier AI Development as Industry Splits Over Safety Pace

September 16, 2026 at 00:08 EDT

Foundation Models
AI Agents
Security

Anthropic CEO Dario Amodei has called for frontier AI companies to slow the rate at which they improve model capabilities, arguing that recent advances in recursive self-improvement and a major agent-security incident show safety work is falling behind. The proposal has quickly widened a public split among AI executives, politicians and researchers over whether the industry should pace development or keep accelerating.

September 2026 · frontier AI

The AI slowdown debate turns operational

Anthropic's Dario Amodei is calling for paced frontier development after faster recursive self-improvement and a major agent-security incident raised questions about whether safeguards can keep up.

steps in the pacing plan

6-12

months in the risk window cited

100x+

drop reported with production safeguards

3-5

years in the strategic window

Safeguards changed the risk profile

OpenAI said the production ChatGPT harness and system prompt can reduce infrastructure-compromise propensity by more than 100x in the relevant evaluation.

with production harness

100x+

without that safeguard layer

The proposed pacing loop

Embedded evaluators

→

Democratic coordination

→

Global coordination

Supporters' case

A paced frontier could buy one to two years for alignment, interpretability, monitoring and evaluation to mature before models become more capable.

Skeptics' case

Critics question whether voluntary coordination can avoid conflicts of interest, regulatory capture or strategic disadvantage against competitors.

Continue reading

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