Google, OpenAI, and Anthropic are building their own AI regulator
The three leading frontier labs are creating FASA, an industry self-regulatory body modelled on FINRA, targeting launch by end-2026 or early 2027.
What is FASA and why does it exist?
Google, OpenAI, and Anthropic are co-founding a new industry self-regulatory body called the Frontier AI Standards Agency (FASA). First reported by The Information, the organisation is designed to fill the oversight gap left by governments that have so far failed to establish a credible, well-resourced regulator for frontier AI models. The target launch window is either end of 2026 or early 2027.
The idea originated with Google DeepMind co-founder Demis Hassabis, who outlined a proposal on 14 July 2026 in an essay calling for an independent standards body modelled on FINRA, the financial industry’s self-regulator. That framing appears to have unlocked rare agreement across competing labs. OpenAI Chief Global Affairs Officer Chris Lehane confirmed on 15 September 2026, at a Washington briefing, that the three labs had been coordinating on safety protocols for several weeks.
Who is being asked to run it?
The three companies have approached Sriram Krishnan to serve as FASA’s chief executive. Krishnan served as the Trump administration’s Senior White House Policy Adviser on Artificial Intelligence until June 2026. During his 18 months in that role, he was an outspoken opponent of a licensing regime for AI, saying publicly that “there will not be an FDA for AI.” His argument was that centralised oversight would slow progress unnecessarily.
The irony is not subtle. The person most publicly associated with the position that a frontier AI regulator should not exist is now being asked to lead one. Whether that makes him the ideal bridge builder or a walking contradiction depends on who you ask.
Other candidates reportedly approached for leadership or advisory roles include former Biden administration science adviser Arati Prabhakar, former Secretary of State Condoleezza Rice, and venture capitalist David Friedberg. On the technical side, the labs have reportedly courted Beth Barnes, founder and CEO of METR, and Paul Christiano, who advises the Center for AI Standards and Innovation.
What will FASA actually do?
The proposed body is modelled explicitly on FINRA, the Financial Industry Regulatory Authority that oversees broker-dealers in the United States. The functions being discussed include:
- Pre-deployment safety testing, with labs sharing models for review up to 30 days before release
- Third-party evaluation frameworks covering cybersecurity vulnerabilities, biological threat potential, and agentic behaviours such as guardrail bypass and model-level deception
- Incident reporting rules
- Qualification standards for auditors
The design is voluntary at launch, with the intention to transition to mandatory protocols as the body builds credibility. Notably, FASA would not require approval from Congress or the White House to operate, though it would likely need to register with a government agency if it ever wanted to wield formal legal authority.
This is a meaningful structural shift from the Frontier Model Forum, which Google, Microsoft, OpenAI, and Anthropic formed in 2023 with a $10 million safety fund. That body has not stopped a single release, required a single evaluation, or enforced a single standard. FASA is being positioned as more formal, with auditor certifications and pre-release requirements.
How did we get here?
The original plan was different. The three labs initially sought to establish a public-private partnership under federal oversight. That path closed when a draft White House executive order was put on hold, prompting a shift to the industry-led model now being assembled.
Safety concerns have also become harder to dismiss. Agentic deployments have produced real incidents, including a breach in which OpenAI agents accessed third-party systems via Hugging Face infrastructure. On 23 September 2026, OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei, and Hugging Face CEO Clem Delangue addressed the United Nations Security Council on AI safety. Earlier in September, Amodei published an essay calling for an industry-wide slowdown in frontier model releases to allow safety mechanisms to catch up, which accelerated momentum toward collective action.
On the government side, Bloomberg has reported that Treasury Secretary Scott Bessent helped develop a separate proposal, now under review by White House Chief of Staff Susie Wiles, for an independent frontier AI regulator reporting to the SEC. Whether that becomes a parallel track or a competitor to FASA remains to be seen.
The criticism you should take seriously
Cohere CEO Aidan Gomez did not mince words. On 14 September 2026, he called the proposal “a cartel by any other name”, drawing comparisons to the SEC’s 1975 NRSRO designation and Europe’s 1985 Motor Vehicle Block Exemption, both of which entrenched incumbents while raising barriers for smaller players.
His argument is worth sitting with. The coordination that would ordinarily raise antitrust flags becomes acceptable once it is labelled safety cooperation. If FASA standardises evaluation frameworks that only the three founding labs helped design, every other developer, including open-source projects and smaller commercial competitors, would be expected to comply with standards they had no hand in writing. Gomez’s framing: rebranding a competitive advantage as a public good.
The structural conflict applies here too. FASA is being funded and staffed, at least in part, by the same companies whose models it will evaluate. That is not unique to AI. FINRA has the same tension. But it is worth naming clearly rather than assuming the problem is solved by good intentions.
There is also a practical bottleneck. When METR and Redwood Research evaluated an OpenAI agent deployment, they spent six days on-site and received the full dataset only in the final two days. Apollo Research had three days with GPT-6 Astra, only two of which included chain-of-thought access. If FASA is going to run meaningful pre-deployment evaluations, the resourcing question is not academic.
What does this mean for you?
If you are an enterprise technology leader, the most immediate implication is uncertainty in your release planning. Sam Altman told the G20 Innovation Ministerial that next-generation models will be “sobering for everybody.” If FASA standardises pre-release evaluation windows, the 30-day review period alone will stretch release cadences that enterprise roadmaps have been built around.
If you are a developer outside the three founding labs, pay attention to Gomez’s critique. The audit and certification standards being written now will shape what compliant AI development looks like for years. There is currently no confirmed mechanism for smaller developers or open-source contributors to influence those standards.
If you are watching this from a policy perspective, the key signal to watch is whether Krishnan accepts the role, and whether Anthropic and Google each publicly confirm participation on their own terms rather than deferring to OpenAI’s framing. A founding charter or CEO announcement before the end of 2026 would confirm this is a live regulatory effort. Silence into 2027 would suggest it is still a proposal looking for a structure.
The gap in government oversight is real. The U.S. Center for AI Standards and Innovation exists but insiders acknowledge it lacks the technical depth and financial resources to evaluate complex frontier systems at scale. Something is going to fill that gap. The question worth asking is whether the organisations being evaluated should be the ones building the body that evaluates them.