Google, OpenAI, and Anthropic are teaming up to create their own AI safety standards body — no government oversight involved.
The proposed organization, tentatively called the Standards Authority for Frontier AI (SAFA), aims to launch by late 2026 or early 2027, focusing on testing and auditing standards for frontier AI models.
How this started
Back in July, Google DeepMind CEO Demis Hassabis publicly proposed that the U.S. should establish an AI self-regulatory organization modeled after FINRA (the Financial Industry Regulatory Authority), rather than creating a new federal agency.
That idea got the ball rolling. Representatives from Google, OpenAI, and Anthropic began meeting regularly as a working group. Initially, they tried to set up a public-private partnership with federal oversight, but it stalled — the industry couldn't reach consensus, and the government wasn't interested. A White House executive order draft on the matter was quietly shelved before summer ended.
With government action off the table, they decided to go it alone.
According to insiders, the three companies have approached Sriram Krishnan for the CEO role. Krishnan, who served as the White House's senior AI policy advisor from January 2025 to June 2026, has both the experience and connections. SAFA's focus will be on concrete testing and auditing frameworks, not vague ethical guidelines. Models would need to pass third-party safety testing before deployment, and there would be incident reporting requirements.
Why big companies want to write their own rules
Three reasons.
First, government regulation is slow and often too broad. AI evolves fast — by the time legislation passes, the technology may have already gone through three generations. Better to set your own rules than have someone else set them for you.
Second, compliance costs become competitive barriers. Once SAFA standards are in place, smaller companies will need to invest heavily to meet them. The OECD has already warned that high fixed costs are making it difficult for new players to enter the AI market. Another layer of compliance costs only raises the bar higher.
Third, trust. The UN report was blunt: AI risk is no longer theoretical. Waiting for a disaster before taking action isn't an option. SAFA is positioning itself as a preventive measure, not a cleanup crew.
What this means for brands
This might seem like an internal AI industry matter, but it has direct implications for every brand.
Think about it: when AI search, AI recommendations, and AI agents become the primary way users get information, which sources those AI systems "trust, cite, and recommend" becomes a standards issue.
GEO (Generative Engine Optimization) is precisely about getting your brand onto AI's trusted list. The stricter the AI safety standards, the more selective AI systems become about information sources. Brands with high-quality content, reliable data, and traceable sources will naturally be favored. Those relying on content farms, keyword stuffing, and manipulative tactics will find themselves squeezed out.
Once SAFA's testing and auditing standards are established, they'll indirectly influence how AI systems evaluate and select information sources. In other words, when the "referee's" standards go up, the "players" need to level up too.
Big companies set the rules — what about everyone else
Here's the concern: Google, OpenAI, and Anthropic are both AI model developers and the ones writing the safety standards. Grading your own homework doesn't exactly inspire confidence.
Analysts point out that SAFA needs broader industry buy-in and clear boundaries with government regulators, otherwise it's just a "big boys' club."
But the trend is unmistakable: the AI industry is moving from "wild west" to "rulebook." For brands, rather than waiting for regulations to force adaptation, now is the time to solidify your content foundation — accurate data, clear sourcing, substantive viewpoints.
That's the strongest hand you can hold in the GEO era.