Breaking

Anthropic's Amodei Calls for Slower Frontier AI Development as Rivals Signal Support

September 14, 2026 at 16:06 EDT

Foundation Models
AI Agents
Security

Anthropic CEO Dario Amodei has called for frontier AI developers to slow capability gains and add embedded outside evaluators, arguing that safety systems are falling behind agentic models. The proposal drew support from OpenAI's Sam Altman, Elon Musk and Google DeepMind's Demis Hassabis, while critics warned that coordinated pacing could entrench incumbents or weaken U.S. competition with China.

September 2026 · Frontier AI

AI Labs Hit The Brakes

Anthropic is calling for slower frontier model development, embedded outside evaluators and coordination as agentic systems begin to accelerate AI research itself.

6-12

month risk window cited for more capable agent swarms

80%+

of Anthropic merged code authored by Claude

more code per engineer per day than in 2024

week pause in OpenAI RL training on latest deployment-bound models

The productivity signal behind the safety debate

1x2024 baseline

8xQ2 2026

AI-written code is speeding up model work, but the same acceleration raises the cost of weak monitoring.

Pacing proposal

Embedded evaluators

Outside reviewers with employee-like access.

→

Shared standards

Common safety checkpoints among frontier labs.

→

Global coordination

Verification before broader pauses can be credible.

Supporters see a safety window

More capable agents can code, coordinate and probe systems; independent review could help safeguards keep pace.

Critics see a competition risk

Coordinated slowdown could favor incumbents, burden smaller labs or weaken the U.S. lead if rivals do not follow.

Continue reading

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