OpenAI, Google, and Anthropic in talks on shared AI safety standards

The three organizations at the frontier of large language model development operate under no common safety standard.

$Marisol Quintero·NewsNTech·[GOOGL]

share //

Key takeaways

OpenAI, Google, and Anthropic are in active discussions about collaborating on AI safety issues, confirmed by a spokesperson.
The three organizations at the frontier of large language model development currently operate under no common safety standard, each setting its own internal deployment thresholds.
The talks have been running since July, when Google's Demis Hassabis released a proposal calling for a U.S.-led "Standards Body" for the industry.
The proposed Standards Body would position an external institution above any individual lab's internal review process, differing from informal peer coordination.
The specific shape any agreement might take has not been disclosed.

The three organizations at the frontier of large language model development operate under no common safety standard. Each lab sets its own internal thresholds for what constitutes a safe deployment, and that gap has become harder to ignore when the labs involved are also the principal competitors at the current capability tier. A spokesperson confirmed that OpenAI, Google, and Anthropic are in active discussions about collaborating on AI safety issues.

The talks have been running since July, when Google's Demis Hassabis released a proposal calling for a U.S.-led "Standards Body" for the industry. The concept positions an external institution above any individual lab's internal review process. That is a different structural arrangement than informal peer coordination between competitors.

The spokesperson's confirmation is the current state of play. The specific shape any agreement might take has not been disclosed.

$ cat source.txt

cnbc.com

← back to the feed Subscribe free

// related, decompiled

Nvidia's Jensen Huang breaks from Anthropic and OpenAI on AI safety at Dreamforce

The pace of AI model development has become the sector's central safety argument, and at Dreamforce that argument broke open publicly. Nvidia chief executive Jensen Huang diverged from both

Devon Ashkar·1 minSEP 15

[TECH]02

Jensen Huang calls AI safety-versus-innovation debate a 'false choice'

The governance debate inside AI runs on a contested premise: that advancing capability development and maintaining responsible safety controls are, at some level of speed, incompatible.

Devon Ashkar·1 minSEP 15

[TECH]03

AI rehearsal tools move into the manager's pre-meeting prep stack

The bottleneck in high-stakes workplace conversations has always been preparation time and the inability to stress-test a response before it matters. AI tools are now being used by managers to rehearse difficult

Devon Ashkar·2 minSEP 13

Frequently asked

Which companies are involved in the AI safety standards talks?

OpenAI, Google, and Anthropic, the three organizations at the frontier of large language model development, are involved in the discussions.

When did the talks begin?

The talks have been running since July, when Google's Demis Hassabis released his proposal.

What did Demis Hassabis propose?

He released a proposal calling for a U.S.-led "Standards Body" for the industry, positioning an external institution above any individual lab's internal review process.

Has a final agreement been reached?

No, a spokesperson only confirmed active discussions, and the specific shape any agreement might take has not been disclosed.

Why are shared safety standards being discussed now?

The labs currently operate under no common safety standard, and that gap has become harder to ignore because the labs involved are also the principal competitors at the current capability tier.