Short Answer: What is the Standards Authority for Frontier AI?
Google, OpenAI, and Anthropic have reportedly agreed to jointly establish the Standards Authority for Frontier AI (SAFA), an independent, industry-funded body that would set common benchmarks and test frontier AI models before release. Modeled on the Financial Industry Regulatory Authority (FINRA), it could launch by the end of 2026 or in early 2027, and the three labs have approached several high-profile figures to help run it.
Quick Summary
- The Information reported that Google, OpenAI, and Anthropic have agreed to establish the Standards Authority for Frontier AI, or SAFA, an independent third-party body to test and monitor frontier AI models.
- The idea originated from a July 14, 2026 proposal by Google DeepMind CEO Demis Hassabis, calling for a self-regulating “standards body” modeled on FINRA, Wall Street’s self-regulatory organization.
- The three companies have approached former White House AI policy adviser Sriram Krishnan, former Biden administration technology official Arati Prabhakar, former diplomat Condoleezza Rice, and venture capitalist David Friedberg for leadership roles.
- Potential scientific consultants include METR founder and CEO Beth Barnes and Paul Christiano, an adviser to the Center for AI Standards and Innovation.
- The proposal follows a wave of AI executives, including OpenAI’s Sam Altman and Anthropic’s Dario Amodei, telling the United Nations Security Council on September 23 that the industry needs stronger oversight.
The Information reported that Google, OpenAI, and Anthropic have agreed to establish a joint standards organization for frontier AI models, tentatively called the Standards Authority for Frontier AI, or SAFA. The body would act as an independent third party that tests advanced models and sets common benchmarks for the industry, and could launch by the end of 2026 or in early 2027.
Where Did This Idea Come From?
The proposal traces back to a July 14, 2026 essay by Google DeepMind CEO Demis Hassabis, titled “A Framework for Frontier AI and the Dawning of a New Age.” Hassabis called for a “standards body” modeled directly on the Financial Industry Regulatory Authority (FINRA), the organization that oversees U.S. broker-dealers on Wall Street.
Under his proposal, frontier labs would initially share models with the standards body voluntarily, up to 30 days before release, for testing in high-risk domains like cybersecurity and biological threats, including agentic tests for attempts to bypass safety guardrails or signs of deception. “Once the assessment protocol is shown to be effective and robust, formalisation could quickly follow,” Hassabis wrote, “meaning that Frontier Models would be required to pass it to be deployed in the US market.”
Hassabis argued the approach would stay “technically focused, while at the same time supporting innovation and incentivising responsible behaviour,” and that it could adapt and tighten if risks grow more serious. The idea drew notably broad, quick agreement across a normally competitive industry: OpenAI CEO Sam Altman called it “thoughtful,” Elon Musk called it “a thoughtful framework overall and certainly a good starting point for discussions,” and former White House AI czar David Sacks said he thought the idea had merit.
Who Would Run It?
According to the Information’s reporting, the three companies have approached several high-profile figures for leadership positions at SAFA:
- Sriram Krishnan — former Senior White House Policy Advisor on AI, who left the second Trump administration in June 2026.
- Arati Prabhakar — a former Biden administration technology official.
- Condoleezza Rice — former U.S. Secretary of State and George W. Bush-era diplomat.
- David Friedberg — venture capitalist.
Potential scientific consultants reportedly approached include Beth Barnes, founder and CEO of METR, the independent research group that investigated the Hugging Face agent-breach incident, and Paul Christiano, who advises the Center for AI Standards and Innovation and was recently appointed to OpenAI’s foundation board. Both have had recent dealings with OpenAI, which the reporting notes as relevant context for their potential involvement.
One detail worth flagging: Krishnan is on record from his time in the White House arguing against a government AI regulator, saying “there will not be an FDA for AI” and warning that a centralized agency requiring “a team of lawyers before you can get a model out” would slow the industry down. His potential involvement in an industry-run alternative is consistent with that position: self-regulation instead of a government body.
How Would It Actually Work?
As a self-regulatory organization, SAFA would not need approval from Congress or the president to operate, and it would set its own testing standards and benchmarks. But if it wants any binding legal authority, it may eventually need to register with a government agency, the way FINRA is registered with and overseen by the Securities and Exchange Commission.
It is not yet clear how SAFA would be funded, though Hassabis’s original proposal argued that funding “would need to be substantial and likely mostly come from industry” in order to attract top technical talent and secure the computing resources needed to actually test frontier models. The body would build on, and could eventually replace, the ad hoc pre-release reviews the U.S. government has already conducted on models like Anthropic’s Mythos and OpenAI’s Sol — reviews that drew criticism for limited technical expertise and opaque decision-making about when a model could ship.
Why Now?
The plan surfaces amid a run of agent-breach disclosures across the industry this year, including incidents involving OpenAI and, most recently, Google’s Gemini, alongside researcher warnings that advanced AI could pose serious risks if development continues to outpace safety work. On September 23, the day before this report, AI executives including Altman, Amodei, and Hugging Face CEO Clem Delangue spoke to the United Nations Security Council specifically about AI safety.
It also follows a broader, unusually consistent message from the major labs this year that AI development should be “paced” so that safety practices and oversight can keep up with capability gains — the same argument Anthropic made around its own R&D automation transparency report and its Claude Opus 5.5 launch earlier this week. Not everyone in the industry agrees with that framing: Nvidia CEO Jensen Huang has said he holds a more accelerationist view.
Most AI-specific regulation in the U.S. currently sits at the state level. Congress has several federal AI bills in committee, none of which have advanced, and while labs can voluntarily submit frontier models to the government’s AI safety institute for pre-deployment testing under a June 2026 executive order, the White House has not published clear guidelines for how that process works. President Trump has separately said he does not believe additional AI regulation is needed.
Why This Matters
Whatever shape SAFA ultimately takes, it represents an unusual level of coordination between three companies that compete directly for the same customers, talent, and model-release headlines. That the proposal moved from a single CEO’s essay to a jointly funded organization with candidate leadership in about ten weeks says something about how urgently the labs themselves see the oversight gap. Whether an industry-funded, self-regulatory body can credibly test its own funders’ most advanced systems is the open question, and it is the same tension FINRA itself has navigated for decades on Wall Street. For now, SAFA remains a plan rather than a launched organization, with real decisions about funding, authority, and independence still ahead of it.
Frequently Asked Questions
Is the Standards Authority for Frontier AI a government agency?
No. As currently proposed, it would be a self-regulatory organization funded primarily by the AI industry, similar in structure to FINRA, which oversees Wall Street broker-dealers. It would not require congressional or presidential approval to operate, though it may eventually need to register with a government agency to hold binding legal authority.
Who came up with the idea for SAFA?
The proposal originated with Google DeepMind CEO Demis Hassabis in a July 14, 2026 essay calling for a FINRA-style standards body to test frontier AI models before release. Google, OpenAI, and Anthropic have since reportedly agreed to jointly establish the organization.
When would SAFA launch?
According to The Information’s reporting, SAFA could launch by the end of 2026 or in early 2027, though the organization is still in the planning stage and key details like funding and legal authority remain unresolved.
Would AI labs be required to submit models to SAFA?
Not initially. Under Hassabis’s original proposal, frontier labs would voluntarily share models for review up to 30 days before release. He suggested that requirement could later become mandatory for deployment in the U.S. market once the testing protocol proved effective.
Do all AI industry leaders support this approach?
Most major AI executives, including Sam Altman and Elon Musk, responded positively to the initial proposal. Not everyone agrees with the underlying premise that development should be paced for safety, however — Nvidia CEO Jensen Huang has publicly taken a more accelerationist position.
Sources: BankInfoSecurity, TechCrunch.
About the author
Kai Williams
Kai Williams has been in marketing for years, with a long background in SEO before AEO had a name. He stepped into Answer Engine Optimization the moment AI started reshaping how people search, and has been tracking the shift ever since. At Prompt Insider, he covers AEO, AI marketing, and the future of search, breaking down what is changing and what brands need to do about it.
