# AI rivals start talking safety

> OpenAI, Google and Anthropic are quietly comparing notes on how fast to push their best models.
- Title: AI rivals start talking safety · CNBC | Shortify
- Summary: OpenAI, Google and Anthropic are quietly comparing notes on how fast to push their best models. The three biggest labs are trying to set their own safety…
- Keywords: ai safety, openai, anthropic, google deepmind, regulation, business, finance, rivals, start, talking, safety, CNBC
- Source: CNBC — https://cnbc.com/2026/09/15/open-ai-google-anthropic-safety.html
- Author: Ashley Capoot
- Published: 2026-09-15T18:31:39+00:00
- Read time: 1 min
- Topics: ai safety, openai, anthropic, google deepmind, regulation, business, finance
## Rivals open a back channel
OpenAI confirmed to CNBC it has been talking with Anthropic and Google about working together on safety concerns. These are three companies fighting each other for the same market.
## It started with Hassabis
Talks have run since July, when Google DeepMind chair Demis Hassabis published a proposal for a U.S.-led standards body for AI.
## Modeled on Wall Street
Hassabis floated a public-private partnership or self-regulatory group with federal oversight, similar to FINRA, the body that polices brokers.
## Amodei calls for a slowdown
On Saturday, Anthropic CEO Dario Amodei urged labs to slow how fast they improve their most advanced models. Hassabis and Sam Altman both backed him publicly.

> Dario's essay points towards the right path forward
## Altman says more soon
Altman agreed on pacing the frontier models and said a slowdown has been a primary topic inside OpenAI in recent weeks, with an announcement coming.
## Voluntary, with or without D.C.
OpenAI policy chief Chris Lehane said the company will build industry-led standards even without government support, framing them as a complement to federal rules, not a substitute.

> everything should be on the table
## Key takeaway

The three biggest labs are trying to set their own safety rules before anyone forces them to.