Have you ever wondered: who will ensure that the most powerful AI machines do not spiral out of control? As large language models become increasingly intelligent, the question of safety is no longer just a topic for researchers. It is a question for society as a whole.
And it seems that the three biggest names in the industry—Google, OpenAI, and Anthropic—are preparing to provide their answer. Not through a generic statement, but through an independent AI safety organization tentatively named: Standards Authority for Frontier AI (SAFA).
This is not just an ordinary cooperative agreement. This is the first time that three companies fiercely competing in the AI market have sat down at the same table to build a common framework—a framework that could shape how AI is developed, evaluated, and deployed in the years to come.
What is SAFA and why does it matter?
According to sources close to the matter, SAFA is expected to launch in late 2026 or early 2027, operating independently from the government. Its primary mission is to shape practical standards for the public safety commitments of the three companies, while providing direct support to developers in evaluating advanced models prior to commercial deployment.
Put simply: SAFA will act as a "gatekeeper" for next-generation AI models. Before a model is released to the market, it will have to pass tests set by SAFA—not by the government, but by an independent organization backed by the industry.
This raises an interesting question: Can an organization funded by AI companies themselves truly be independent in its evaluations? Or is this just a way for the industry to regulate itself without state intervention?
The figures behind SAFA
To better understand SAFA's direction, let's look at the figures rumored to lead this organization.
The three companies are reported to have approached Sriram Krishnan, a former White House AI policy advisor, to take on the role of chief executive officer. They have also contacted Arati Prabhakar, former director of the White House Office of Science and Technology Policy, for a senior leadership role at SAFA.
Both individuals have government backgrounds, but SAFA is designed to operate outside the state apparatus. This reflects a clear philosophy: the industry wants to shape its own standards rather than wait for external regulations.
The three companies are also considering whether SAFA should conduct model evaluations directly in-house. This is a crucial decision, because if SAFA evaluates models itself, it will need to build substantial technical capacity—something that even the U.S. National Artificial Intelligence Safety Institute (CAISI) reportedly lacks in both technical and financial resources to test complex advanced systems.
Context: Why now?
This move does not happen in a vacuum. It is the result of a chain of events and mounting pressure within the industry.
Proposal from Demis Hassabis
In July 2026, Sir Demis Hassabis, co-founder of Google DeepMind and Chief Scientist at Alphabet, published a proposal on X. He called for the establishment of a U.S.-backed standards body, modeled after a public-private partnership or an industry self-regulatory organization.
Hassabis compared this model to FINRA—the Financial Industry Regulatory Authority, a self-regulatory organization overseeing brokerage firms and securities markets in the U.S. This is a notable comparison: FINRA operates independently but under the oversight of the U.S. Securities and Exchange Commission (SEC).
Essay by Dario Amodei
In early September 2026, Anthropic CEO Dario Amodei published an essay calling for a slowdown in the industry regarding the release of advanced models. He argued that time is needed for safety mechanisms to catch up with the pace of technological development.
This is a stunning statement, especially coming from the CEO of one of the companies leading the AI race. It shows that even insiders are aware of the potential risks of developing unchecked AI.
Public and governmental pressure
All three companies are facing increasing scrutiny from the public, media, and lawmakers. Concerns about catastrophic risks and security associated with advanced AI have become hot topics in policy discussions.
In that context, voluntarily establishing an independent safety organization can be seen as a way to reduce regulatory pressure—a strategy of "self-regulate before being regulated."
Challenges and unresolved questions
While the idea of SAFA sounds promising, many questions remain unanswered.
First, the true independence of SAFA. If this organization is funded by the very companies it evaluates, can it deliver objective assessments? Will SAFA dare to say "no" to a model developed by Google or OpenAI if it fails to meet safety standards?
Second, the enforcement mechanism. What powers will SAFA have? If a company refuses to comply with SAFA's standards, what will happen? Without a clear enforcement mechanism, SAFA might just become an advisory organization without any teeth.
Third, the relationship with the government. SAFA is designed to operate independently of the government, but is that feasible in practice? AI regulations are being developed in many countries, and an independent organization will likely have to interact with state regulatory bodies in complex ways.
Fourth, the scope of application. Will SAFA apply only to the three founding companies, or will it expand industry-wide? If limited to just three companies, will it have enough power to create systemic change?
Implications for Vietnam and Southeast Asia
For Vietnam and Southeast Asia countries, the development of SAFA brings both opportunities and challenges.
Regarding opportunities: If SAFA succeeds in establishing clear safety standards, AI developers in Vietnam will have a reference framework to ensure their products meet international requirements. This is especially important for companies looking to export AI solutions to the global market.
Regarding challenges: Standards set by SAFA could become barriers to entry for smaller companies, including those in Southeast Asia. If compliance requires significant technical and financial resources, AI startups in the region might be left behind.
Regarding governance: The birth of SAFA raises questions about the role of nations in managing AI. Should Vietnam participate in similar initiatives at the regional level? Do we need our own AI safety organization tailored to our context and priorities?
When rivals become allies
I have witnessed many waves of technology over the past two decades, but I have never seen anything quite like this. Three companies fiercely competing for the top spot in the AI race—Google, OpenAI, and Anthropic—are sitting at the same table to build a shared safety framework.
That speaks volumes. It shows that even insiders recognize that the pace of AI development is outpacing our ability to control it. And without coordinated measures, the consequences could be disastrous.
Yet I cannot help feeling skeptical. Can a company-funded organization truly be independent? Will SAFA become an effective self-regulatory mechanism, or just a way for the industry to avoid government intervention?
Time will tell. But one thing is certain: the conversation around AI safety has entered a new phase. And all of us—whether users, developers, or policymakers—have a responsibility to monitor and participate in that conversation.
What do you think about the world's three largest AI companies cooperating to establish an independent safety organization—is this a welcome pioneering step, or just a strategy for the industry to self-regulate without government oversight? If SAFA is genuinely launched in 2027, what do you expect from it—strict and transparent standards, or loose regulations designed not to hinder innovation? And for Vietnam, should we proactively participate in international AI safety initiatives, or build our own framework suited to the domestic context? Share your views below, because in this AI revolution, silence is no longer a safe option!
0 Comment
Add your comment to this article