The proposed Standards Authority for Frontier AI would let the biggest labs write their own safety rules — and critics are already calling it a cartel.
Image: Illustration of three AI figures meeting around a conference table with a safety shield emblem. Credit: Red Eye Eagle News / original AI illustration.
September 25, 2026 — Google, OpenAI and Anthropic are reportedly working together to create an independent industry body for frontier AI safety, according to a report published September 24, 2026 by The Information. The effort is targeting a launch in late 2026 or early 2027.
Editor's Highlight: The key insight is who holds the pen: this proposal would put the world's most powerful AI labs in charge of writing and enforcing their own safety rules, with no government oversight — which is exactly why rival developers are warning it could function as a cartel.
Table of Contents
- Key Facts
- What the reports describe
- Why the labs want to set their own rules
- What the new body would actually do
- Who might run it
- The cartel criticism
- How this differs from the Frontier Model Forum
- Global context: AI takes the UN stage
- Timeline: what happens next
- Bottom Line
- FAQ
- Sources
Key Facts
| Detail | Information |
| Story | Reported plan for an independent AI safety body |
| Companies involved | Google, OpenAI, Anthropic |
| Proposed name | Standards Authority for Frontier AI (SAFA) |
| Reported launch target | Late 2026 or early 2027 |
| Government role | None — industry self-regulation |
| Key functions | Pre-deployment testing standards, incident reporting, auditor qualifications |
| Main criticism | Could box out open-source developers and competitors |
| Origin of idea | Demis Hassabis, July 2026, FINRA-inspired |
What the reports describe
According to reports, the three labs are developing a new organization tentatively called the Standards Authority for Frontier AI, or SAFA.
The body would be independent, with no government oversight. According to the reporting, the companies initially wanted a public-private partnership, but that effort stalled inside the Trump administration.
Newsroom Focus: Treat SAFA as a reported proposal, not a finalized launch. The Information's September 24 report describes an industry working group, and none of the three companies has formally announced the body.
Why the labs want to set their own rules
The idea originated in July with Google DeepMind co-founder Demis Hassabis, who was inspired by Wall Street's Financial Industry Regulatory Authority (FINRA) — the self-regulatory body that polices broker-dealers.
The timing is no accident. On September 23, Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman addressed a UN Security Council meeting on AI, where the two rivals agreed on the need for global standards.
OpenAI said in a blog post this week that international standards "may be as important to pacing the frontier as alignment research itself." The company also said that fully autonomous recursive self-improvement is not happening today.
Analysis Note: The FINRA comparison is revealing. Self-regulatory organizations can move faster than governments, but they raise capture concerns — the industry being regulated writes the rules and picks the referees.
What the new body would actually do
According to the reporting, SAFA's planned functions would include:
- Supporting third-party organizations that test models before deployment.
- Setting protocols for reporting AI safety and security incidents.
- Defining the labs' voluntary safety and security commitments.
- Establishing qualifications for independent auditors of models and labs.
The working group is also debating whether the body itself should directly test models' safety and capabilities — a step that would place it alongside federal efforts such as the Center for AI Standards and Innovation.
Practical Takeaway: If you build on frontier models, watch two open questions: whether SAFA tests models directly, and how auditor qualifications are defined. Those decisions could shape compliance costs across the AI industry.
Who might run it
The consortium has reportedly approached former White House AI policy adviser Sriram Krishnan and other notable figures about executive leadership roles.
No leadership appointments have been announced. The body's governance structure remains under discussion inside the working group.
The cartel criticism
Critics — including competing AI developers — worry the three companies could use the body to box out open-source model developers and other competitors.
The concern is straightforward: if the biggest labs define what counts as "safe," the compliance bar could be set at a height only the biggest labs can clear.
That tension sits at the heart of the story. The labs frame self-regulation as speed and expertise; rivals hear gatekeeping.
How this differs from the Frontier Model Forum
A 2023 nonprofit with similar goals, the Frontier Model Forum — founded by Google, Microsoft, OpenAI and Anthropic — remains active.
The two groups are reportedly in contact and could work together, according to the reporting.
| Aspect | SAFA (reported proposal) | Frontier Model Forum |
| Founded | Targeted late 2026 / early 2027 | 2023 |
| Companies | Google, OpenAI, Anthropic | Google, Microsoft, OpenAI, Anthropic |
| Status | Working group stage | Active nonprofit |
Global context: AI takes the UN stage
The SAFA reports landed one day after AI's biggest names took the world stage. On September 23, Dario Amodei and Sam Altman addressed the UN Security Council on AI and agreed on the need for global standards.
Microsoft vice chair Brad Smith backed independent evaluators and keeping a human "off switch" on the most advanced systems.
Taken together, the week marks a shift: the frontier labs are no longer just asking for regulation — they are trying to design it themselves.
Timeline: what happens next
| Date | Milestone |
| July 2026 | Idea originates with Demis Hassabis, inspired by FINRA |
| September 23, 2026 | Amodei and Altman address the UN Security Council on AI |
| September 24, 2026 | The Information reports the SAFA plan |
| Late 2026 – early 2027 | Reported launch target for the new body |
Bottom Line
The world's three leading AI labs want to govern themselves before governments govern them. The reported Standards Authority for Frontier AI would set testing standards, incident-reporting protocols and auditor qualifications — all without government oversight.
What to watch next: whether the body will test models directly, who agrees to lead it, and whether open-source developers get a seat at the table or a wall built around them.
FAQ
What is SAFA?
SAFA — the Standards Authority for Frontier AI — is the tentative name for a reported independent industry body for frontier AI safety being developed by Google, OpenAI and Anthropic. It is a proposal described in reporting, not a formally announced organization.
Why do the labs want to regulate themselves instead of waiting for government rules?
According to the reporting, the companies first wanted a public-private partnership, but that effort stalled in the Trump administration. OpenAI has argued this week that international standards "may be as important to pacing the frontier as alignment research itself."
Who is criticizing the plan and why?
Competing AI developers are among the critics. They worry the three companies could use the body to box out open-source model developers and other competitors by setting compliance bars only the biggest labs can clear.
How is this different from the Frontier Model Forum?
The Frontier Model Forum is a 2023 nonprofit founded by Google, Microsoft, OpenAI and Anthropic with similar safety goals, and it remains active. The two groups are reportedly in contact and could work together.
When could SAFA actually launch?
The reported target is late 2026 or early 2027. The effort is still in a working-group stage, with open questions including leadership and whether the body will directly test models.
Sources
- PYMNTS (Sept 24, 2026) — OpenAI, Google and Anthropic Join Forces to Set AI Safety Standards
- CIO — The Companies Racing to Build Frontier AI Are Now Racing to Govern It
- harava.news (Sept 25, 2026) — Frontier AI safety watchdog
- Original AI-generated illustration for Red Eye Eagle News (hero image)
Join the conversation
Comments
Post a Comment