The short version: Google, OpenAI and Anthropic are building an industry body to set safety rules for the most powerful AI models, and on September 24, 2026, The Information reported that the three have approached former White House AI adviser Sriram Krishnan to run it. The reported working name is the Frontier AI Standards Agency, and the stated model is FINRA, the self-regulator that oversees Wall Street brokers. The comparison holds up except at one point, and it is the point that matters. FINRA has force because federal law requires brokers to join it before they can do business, and it answers to the SEC. Nothing reported so far gives the AI body either one. For a small business renting ChatGPT, Claude or Gemini, that is the question worth watching: a standard your vendor can walk away from protects you only for as long as the vendor stays in.
What is the Frontier AI Standards Agency?
So far it is a report, not an announcement. The Information’s story is paywalled, so the details here come from summaries published by AI Weekly and CASRAI, and we are naming them as our sources. None of the three companies has published a charter.
According to those summaries, staff from the three labs have met as a working group since July, after Google DeepMind’s Demis Hassabis proposed a standards body that month. The functions under discussion include safety testing by outside parties before a model is released, incident-reporting requirements, published model cards, and testing coordinated with federal agencies and national labs on national security questions. A charter or a chief executive is expected before the end of the year. Krishnan has been approached; nobody has reported that he accepted. Some coverage uses a second working name, the Standards Authority for Frontier AI.
OpenAI is the only lab that has spoken on the record. At a Washington briefing on September 15, first reported by Bloomberg and quoted by PYMNTS, OpenAI’s policy chief Chris Lehane said, “There are many instances over time where companies are trying to help each other on safety.” He added that OpenAI does not believe it needs an antitrust waiver to coordinate, pointing to cooperation in the airline industry as precedent.
What does “modeled on FINRA” actually mean?
It helps to read what FINRA says about itself. FINRA’s own About page, read September 24, describes it as a self-regulatory organization operating under the oversight of the Securities and Exchange Commission, and says its operations are funded by member fees, not taxpayer dollars.
The part that gives it teeth is in the SEC’s guide to broker-dealer registration, under Section 15(b)(8) of the Securities Exchange Act: before it begins doing business, a broker-dealer must become a member of a self-regulatory organization. Membership is not a pledge. It is the price of being allowed to sell.
So FINRA stands on three legs: membership the law requires, a government agency above it, and funding from the industry it polices. The AI body, as reported, has the third. Membership is voluntary, and no summary describes any government agency with authority over it. A broker that dislikes a FINRA finding still has to answer to FINRA. A lab that dislikes a finding from a voluntary body can leave it.
Hasn’t the industry tried this before?
Yes. In July 2023, OpenAI, Google, Microsoft and Anthropic launched the Frontier Model Forum, as TechCrunch reported at the time. Its website now lists Amazon, Anthropic, Google, Meta, Microsoft and OpenAI as members and describes three jobs: identifying best practices, advancing independent research and sharing information. Nothing on the site describes audits of its members or penalties for breaking a standard.
That context is also what makes the new proposal worth taking seriously. Testing by outside parties before release and required incident reports are enforcement-shaped ideas, not best-practice ideas. If they survive into a charter, this would be a real step past what the Forum does.
Where does Congress fit?
The missing legs could come from law. The FRONTIER Act, H.R. 9925, from Representatives Jay Obernolte and Lori Trahan, would license independent verification organizations to review the largest AI developers and report to a new Under Secretary of Commerce every six months, according to the Foundation for American Innovation’s analysis of the bill. It defines a large frontier developer as one with more than $50 million in revenue and at least $1 billion in development spending. PYMNTS reports that Lehane voiced support for the bill at the same briefing. A private standards body plus licensed verifiers would start to look much more like FINRA than either one does alone.
What does this mean for a small business that uses AI?
Nothing in your contract changes this year. The good news is real, though. Safety testing by outside parties before release is exactly the kind of check a small buyer could never run on their own, and until now they have had to take the vendor’s word for it.
When a charter does appear, two details will tell you whether it helps you or only the labs:
Are incident reports public? FINRA’s most useful tool for ordinary customers is BrokerCheck, a public lookup for finding a broker or financial advisor and checking their record. The AI equivalent would be a public record of when a model failed and what was fixed. An incident report filed privately with the other labs protects the labs. One you can read protects you.
What does it cost a lab to leave? If the answer is nothing, the standard is a promise, and you should read it the way you read any vendor promise.
Keep one more limit in mind. As reported, the body would cover the labs that build frontier models, not the scheduling app, website chatbot or bookkeeping tool built on top of one. Those products inherit whatever testing the underlying model got, and nothing more. We covered the pressure building inside these same labs in the employee letter asking for a brake on AI development, and what safety spending already costs the tools you rent in OpenAI’s safety line item.
Frequently Asked Questions
What is the Frontier AI Standards Agency?
It is the reported working name for an industry safety body being planned by Google, OpenAI and Anthropic. According to summaries of The Information’s September 24, 2026 report, it would cover testing by outside parties before a model is released, incident reporting and published model cards, and the three labs have approached former White House AI adviser Sriram Krishnan to lead it. No charter has been published.
Is the AI standards body like FINRA?
It is modeled on FINRA, but it lacks what gives FINRA force. Under Section 15(b)(8) of the Securities Exchange Act, a broker-dealer must join a self-regulatory organization before doing business, and FINRA operates under SEC oversight. As reported, membership in the AI body would be voluntary, with no government agency above it.
Is this the same as the Frontier Model Forum?
No. The Frontier Model Forum, launched in July 2023, describes its work as identifying best practices, advancing research and sharing information, and its website describes no audits or penalties. The new proposal reportedly adds testing by outside parties before release and incident-reporting requirements.
Does this change anything for a small business using ChatGPT, Claude or Gemini?
Not yet. Nothing changes in your terms this year. When a charter is published, check two things: whether incident reports are public, and whether a lab pays any price for leaving the body.
If the AI tools you rely on had a public failure record like BrokerCheck, would you check it before you signed up?
