Google, OpenAI, and Anthropic are launching a self-regulatory AI standards body called the Frontier AI Standards Agency. Here is what service businesses need to know today.
Ido Cohen · Published 2026-09-24 · AI News
Google, OpenAI, and Anthropic are building their own referee — and they want to write the rulebook before any government does. Today, September 24, 2026, The Information confirmed that the three biggest AI labs have formally approached Sriram Krishnan to lead a proposed self-regulatory body tentatively called the Frontier AI Standards Agency (FASA). For every plumber, dentist, law firm, and HVAC company that has woven ChatGPT, Gemini, or Claude into their business over the past two years, this matters: the organization that sets safety standards for these tools is now taking shape, and it's being designed by the companies that make them.
This story didn't drop from nowhere. According to AI Weekly's September 24 coverage, Google, OpenAI, and Anthropic have been in active working-group talks since July 2026, meeting periodically through the summer. What made today different was the candidate announcement: the three labs have now jointly approached Sriram Krishnan — a former Microsoft and OpenAI executive — to serve as CEO of the new body.
The proposed agency goes by at least two working names across reporting: "Frontier AI Standards Agency" in coverage by The Information and AI Weekly, and "Standards Authority for Frontier AI" (SAFA) in Gurufocus's reporting published hours ago. Both describe the same initiative. Per Gurufocus, the agency "plans to implement specific testing and auditing systems rather than general ethical guidelines," including third-party safety tests before model deployment and standardized incident reporting rules.
This is not a press release. The talks became public in stages, and the structure is still being negotiated. Anthropic CEO Dario Amodei is driving the effort, focused on technical testing and pre-release auditing of frontier models — what Cryptobriefing called "shared protocols for evaluating advanced frontier models, independent safety assessments, and pre-release checks before these systems reach the public." OpenAI's Sam Altman backs an independent, labs-led body in the absence of federal action. Google DeepMind CEO Demis Hassabis proposed the original architecture in July: a self-regulatory structure modeled on FINRA (the Financial Industry Regulatory Authority), the body that governs Wall Street brokers rather than a government agency doing so directly.
The three CEOs still disagree on the exact shape, according to AI Weekly's earlier September 13 reporting. Nothing is finalized. But the directional move is clear: the companies releasing the most powerful AI tools in the world want to set the safety standards for those tools themselves.
You are already regulated downstream whether you know it or not. Every AI tool a service business uses — for writing appointment reminders, generating ad copy, answering customer questions, or summarizing call recordings — is built on a frontier model from one of these three labs. If those models go through a new pre-release audit process, what gets approved, what gets flagged, and what gets delayed directly determines what tools land on your dashboard.
Here's the practical chain:
1. FASA sets a pre-release audit standard for frontier models (GPT-6, Gemini, Claude)
2. Tool vendors (your AI receptionist, your SEO platform, your CRM chatbot) build on those models
3. Audited models get a faster path to market in enterprise and SMB tools; flagged models get delayed or modified
4. Service businesses end up using different tools than they would have — with different capabilities, different data policies, and different risk profiles
The stakes are not hypothetical. All three companies scored C+ or lower on the 2026 Future of Life Institute AI Safety Index, according to Cryptobriefing's reporting. An independent audit body with real teeth could force changes to how these models behave on your customer data — which is good for trust, but could slow feature rollouts.
Here is where I'll be honest: self-regulation by the companies being regulated is a flawed starting point. Cryptobriefing noted the "notable tension" in this initiative given the three companies' own safety scores. AI Weekly's Alexis Dufresne wrote plainly on September 13 that "the three labs racing hardest to build frontier AI are also drafting the referee's rulebook themselves."
The FINRA analogy Hassabis used is instructive — and not entirely reassuring. FINRA works because it operates under SEC oversight and has real enforcement power. The proposed FASA, per current reporting, would launch "without government oversight" according to AI Weekly. That's a meaningful gap.
A tech-insider.org analysis noted that OpenAI confirmed these discussions in September and argued that coordinating on AI safety does not constitute anticompetitive behavior — framing it as "risk management rather than anticompetitive coordination." But critics will fairly point out that three companies controlling the most capable AI models on the planet aligning on which models clear a safety bar could also lock out smaller competitors.
For a service business owner, the honest read is this: a credible, third-party audit standard is better than nothing. If an independent body requires that any model touching customer health records, financial data, or location data passes a specific test before deployment, that protects you. But if the body is effectively a marketing certification run by the same three labs, it provides false assurance — and you should continue doing your own due diligence on any AI vendor.
Based on reporting across The Information, Cryptobriefing, and AI Weekly, here is what FASA/SAFA is expected to focus on:
That last point is the most controversial. Anthropic's Dario Amodei published a September 12 essay titled "We Must Pace the Frontier" arguing for slowing capability development to let safety catch up. OpenAI's chief scientist Jakub Pachocki wrote about coordinated slowdown as "one option on the table" in a September 6 essay. OpenAI even reportedly asked members of Congress whether such coordination would raise antitrust concerns, according to CASRAI's reporting.
If frontier labs actually slow their release cadence under this framework, that could mean fewer new AI features hitting your tools every quarter. For many service business owners who are still catching up to existing capabilities, that might actually be a relief.
The three founding members of this proposed body cover nearly every AI tool a service business is likely to use:
Notably absent: Meta (running its own Muse Spark line), xAI (Grok), DeepSeek, and Mistral. If the FASA standards become industry expectation, those labs face a choice: join and submit to the same audits, or market themselves as the unaudited alternative. For service businesses currently using Meta AI tools for lead generation or ad creative, this is worth watching.
The regulatory backdrop adds urgency. As safer-ai.org noted, California's Transparency in Frontier AI Act (TFAIA), New York's RAISE Act, and Illinois' Frontier AI Safety Act already require large AI developers to manage catastrophic risk through formal frameworks. The EU AI Act's general-purpose AI model provisions entered enforcement in August 2026. FASA would sit alongside, not replace, these rules — potentially making compliance simpler by creating a single standard that satisfies multiple jurisdictions.
You do not need to restructure your business over a proposed agency that hasn't launched yet. But the signal is clear enough to act on:
1. Audit your AI vendor stack. List every AI-powered tool your business uses and identify which underlying model powers it (GPT-6, Gemini, Claude, or something else). You should know this before a standards body or a regulator asks.
2. Read your AI vendors' data policies. If you're using AI tools to handle customer names, medical info, financial data, or location history, check what happens to that data before a model update or audit changes the terms.
3. Do not treat "audited" as "safe." If FASA launches and your AI vendor says their tool passed the new standard, ask what specifically was tested. A pre-release audit for cybersecurity capabilities is different from a privacy audit for SMB customer data.
4. Watch the Meta/xAI split. If FASA creates a two-tier market — audited (Google/OpenAI/Anthropic tools) vs. unaudited (Meta, xAI, open-source) — you'll want a deliberate policy on which tier your business operates in, not a random default.
5. Subscribe to your tool vendors' update emails. When frontier model audits change what a model can or can't do, vendors will update their feature sets. Staying in the loop means you won't wake up one morning to find your AI booking tool has different behavior than last week.
6. Consider this in your agency or consultant conversations. If you work with a marketing agency that uses AI, ask them which models they use and whether they have a process for vetting new AI tools. The standards conversation is your leverage to demand better answers.
What is the Frontier AI Standards Agency (FASA)?
FASA is a proposed industry-led self-regulatory body being formed by Google, OpenAI, and Anthropic to set safety standards for advanced AI models. As of September 24, 2026, the initiative is still in the negotiation phase — no organization has officially launched, no rulebook has been finalized, and the group has approached but not confirmed a CEO. The tentative target is a launch in late 2026 or 2027.
Does this affect the AI tools I'm already using in my business?
Not immediately. The proposed standards body would primarily govern the development and pre-release testing of frontier AI models — things like ChatGPT, Gemini, and Claude. Over time, if pre-release audit requirements change what these models can do or how they handle data, you would see those changes reflected in the tools you use that are built on these models. The timeline for any real-world impact is likely 12-18 months out at minimum.
Why are the AI companies regulating themselves instead of waiting for the government?
The short answer is that government regulation has lagged well behind AI development. According to Cryptobriefing, the three companies "initially sought a public-private partnership with federal oversight, but this approach stalled due to lack of consensus within the AI industry and changing government priorities." OpenAI's Sam Altman has publicly said the lab will proceed with a standards body "with or without government support." Self-regulation is faster than legislation — but also weaker, since it lacks enforcement power backed by law.
What is "frontier AI" and why does the label matter to me?
Frontier AI refers to the most capable AI models at or near the state of the art in reasoning, coding, or handling complex tasks. Current examples include the GPT-6 family, Claude Opus 5.5, and Gemini 3.8. The label matters because FASA's standards would specifically target these high-capability models — not basic automation tools. If your business uses a simple chatbot or templated email tool, it may not be directly affected. If you use tools powered by the latest GPT or Claude models, it is.
What about AI companies not in the group, like Meta or DeepSeek?
Meta, xAI, DeepSeek, and Mistral are not part of the current FASA discussions, based on reporting as of today. This could create a two-tier AI market: tools built on audited models (from Google, OpenAI, Anthropic) vs. tools built on unaudited models. For service businesses, this will eventually become a vendor selection question — do you want tools that carry a safety certification, or tools that move faster but without independent review? Neither answer is automatically correct for every business type.
---
Sources: