Breaking

Anthropic CEO Dario Amodei Calls for Slower Frontier AI Development and Embedded Safety Review

September 23, 2026 at 16:04 EDT

Foundation Models
AI Agents
Policy & Regulation

Anthropic CEO Dario Amodei is urging frontier AI labs to slow the rate at which their models gain new capabilities so safety work can catch up, and Anthropic says it will put independent evaluators inside the company. The proposal is not a call to stop releases; it is a demand for verifiable pacing before more autonomous systems arrive.

Sep 2026 · Anthropic

Pacing the Frontier

Dario Amodei called for frontier AI progress to slow enough for safety work to keep up, with outside evaluators embedded inside Anthropic.

Plan layers: embedded review, democratic coordination, global pacing.

6-12 mo

Warning window for a more capable misaligned agent swarm.

$1B+

Expected investment by each of Anthropic and Accenture over five years.

30,000

Agents doing Anthropic research and engineering work as of August.

AI work inside the lab

Claude's reported role in Anthropic model R&D moved from no led work in February to 26% led work by August, while 90% of R&D involved collaboration.

February led

26%

August led

90%

Collaboration

How the pacing proposal works

Embedded evaluators

Internal access to verify practices and report incidents.

→

Common standards

Democratic-country coordination on safety bars.

→

Global pacing

Narrow agreements where compliance can be checked.

Why supporters like it

They see embedded checks as a way to make safety claims verifiable before more autonomous systems arrive.

Why critics worry

They warn coordinated pacing can create antitrust, competition and regulatory-capture problems.

Continue reading

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