OpenAI, Anthropic and Google DeepMind are discussing the creation of a new industry body to establish AI safety standards, according to reporting from The Washington Post and CNN. The discussions come as AI leaders debate how to manage risks from increasingly capable models and whether companies should coordinate on safety requirements.
The proposed organization would focus on AI safety and could be modeled after the Financial Industry Regulatory Authority, a US financial-industry oversight organization. However, the companies have not publicly established a completed joint regulator, and the exact structure and authority of the proposed body remain unclear.
OpenAI, Anthropic and Google Are Discussing a Shared Safety Organization
The Washington Post reported on September 14, 2026, that Anthropic, OpenAI and Google had been discussing the possible creation of an industry body to set safety standards for AI.
The report, citing three people familiar with the private discussions, said the conversations were taking place before the latest public calls for the industry to slow the pace of AI development.
CNN also reported on September 14 that the three companies had discussed creating an AI industry standards body. According to CNN, the conversations were first reported by The Information.
The key point is that the discussions are about creating a potential organization. They do not establish that the companies have already signed a final agreement or launched a fully operational safety regulator.
The Proposal Behind the AI Safety Body
The idea is connected to a proposal from Google DeepMind’s Demis Hassabis.
In July 2026, Hassabis proposed a US-led standards body that could test the most powerful AI systems before they are released to the public. The proposal was described as similar in concept to the Financial Industry Regulatory Authority, which oversees US brokers and investment firms.
According to RTÉ’s September 15 report, OpenAI said it was working with Anthropic and Google on the idea originating from Hassabis’s proposal.
Hassabis’s proposal called for substantial funding, likely from industry, to support technical expertise and computing resources for large-scale AI testing.
Why AI Safety Standards Are Receiving More Attention
The discussions come during a period of growing concern about the risks associated with advanced AI systems.
Anthropic CEO Dario Amodei recently argued that the pace of AI development needs to slow enough for safety measures to catch up. His proposal included independent evaluators inside AI companies and stronger coordination on AI safety.
OpenAI CEO Sam Altman publicly endorsed aspects of the discussion. Google DeepMind’s Demis Hassabis also expressed support for the broader idea of slowing AI development.
The reporting places the proposed standards body within this wider debate about AI safety, industry competition and the need for more effective safeguards.
What Could the Proposed Organization Actually Do?
The precise responsibilities of the proposed body have not been finalized publicly. However, Hassabis’s earlier proposal gives an indication of the type of work that could be involved.
Testing Advanced AI Models
A central idea is to test powerful AI systems before public release. Such testing could help identify risks associated with increasingly capable models.
The proposal does not establish a final testing protocol, a required benchmark or a specific certification process.
Developing Common Safety Standards
A shared industry body could potentially help companies coordinate their approaches to AI safety. The objective would be to create common expectations rather than having every company operate under entirely separate standards.
Whether the final body would have authority to enforce these standards remains unknown.
Supporting Independent Evaluation
Hassabis’s proposal called for substantial resources to support large-scale testing and technical expertise. Independent evaluation is also part of the wider AI safety discussion.
However, there is no confirmed public agreement showing exactly how independent evaluators would be appointed, funded or governed by the proposed organization.
How Do OpenAI, Anthropic and Google DeepMind Handle AI Safety Today?
The three companies already maintain their own AI safety programs and frameworks.
| Company | Existing safety approach |
|---|---|
| OpenAI | Its Preparedness Framework evaluates certain advanced model risks, including cybersecurity. |
| Anthropic | Its Responsible Scaling Policy and Frontier Compliance Framework address risks associated with advanced AI models. |
| Google DeepMind | Its Frontier Safety Framework uses capability thresholds, evaluations and mitigation plans for severe risks. |
These are separate company-specific systems. They should not be confused with a jointly adopted standard.
OpenAI’s Existing Safety Framework
OpenAI’s September 1, 2026, report on its Astra model said the model met the company’s critical cybersecurity capability threshold under its Preparedness Framework.
The report describes the safeguards OpenAI considered necessary for that capability level.
Anthropic’s Safety Commitments
Anthropic’s Transparency Hub describes its Responsible Scaling Policy, first published in September 2023, as a framework for managing potential catastrophic risks from models.
The company also publishes information about its voluntary safety commitments.
Google DeepMind’s Frontier Safety Framework
Google DeepMind’s Frontier Safety Framework includes processes for identifying capability levels, evaluating risks and preparing mitigation plans.
The framework is designed for the company’s most advanced foundation models.
What This Means for AI Developers and Users
If the proposed standards body is eventually established, it could influence how advanced AI systems are evaluated and how safety expectations are coordinated across companies.
For developers, shared standards could provide more consistent information about model risks and evaluation methods.
For businesses, common safety practices could affect decisions about deploying powerful AI systems and autonomous agents.
For users, the practical effect would depend on what the organization actually does. A discussion about standards does not by itself establish that AI products will become safer, that access restrictions will change or that a common certification system will exist.
What Has Not Been Confirmed?
Several important details remain open.
- The final name of the proposed organization.
- Its official launch date.
- Its membership beyond the companies involved in the discussions.
- Its governance and funding structure.
- Whether its standards would be voluntary or legally enforceable.
- Whether it would independently test models or only publish recommendations.
- Whether governments would participate in its operation.
These details matter because an industry-led standards group would have a different role from a government regulator with legal authority.
The Bigger Question: Who Should Set AI Safety Standards?
The discussions raise a broader question about the future of AI governance.
AI companies are developing their own safety frameworks, while researchers and policymakers are debating whether more coordinated approaches are needed.
An industry standards body could offer a mechanism for collaboration, but its effectiveness would depend on the standards it adopts, the independence of its evaluations and the willingness of companies to follow its recommendations.
For now, the confirmed development is that OpenAI, Anthropic and Google are discussing a possible AI safety standards body. The organization’s final structure, authority and practical impact remain to be established.
Frequently Asked Questions
Q1. Are OpenAI, Anthropic and Google DeepMind creating an AI safety standards body?
They have been reported to be discussing the creation of an AI industry standards body. A completed, operational organization has not been publicly established in the reporting reviewed for this article.
Q2. Who proposed the AI safety standards body?
Google DeepMind’s Demis Hassabis proposed a US-led AI standards body in July 2026. The proposal was described as similar in concept to FINRA.
Q3. Will the new AI safety body regulate AI companies?
Its final authority has not been confirmed. The proposed organization is an industry-led standards initiative, not a confirmed government regulator.
Q4. Why are AI companies discussing shared safety standards?
The discussions come amid concerns about the pace of AI development and the need for stronger safety evaluations and safeguards for advanced AI systems.
Q5. Is the AI safety standards body already operational?
The reporting confirms discussions about creating the body. Its final launch, governance and operating rules remain unconfirmed.
