Imagine three of the world’s largest artificial intelligence companies, Google, OpenAI and Anthropic, sitting down together to write safety rules for AI. At first, this sounds like it could only be good news. But then an uncomfortable question presents itself. Who will control the people writing the rules?
These companies are developing some of the most powerful AI systems in the world, while also proposing the standards by which their own technologies may eventually be assessed.
Meanwhile Anthropic has kicked off with implementing the first of the three phases that head honcho Amodei has laid out as necessary for safe AI development (allowing a team of independent advisors who are external to the company, to have permanent access to data and operations, on a par with workers in the company).
In a way, he has laid down a gauntlet for the others to pick up: will they follow him in this? Time will tell.
Google, OpenAI and Anthropic are currently discussing the creation of a joint organisation provisionally called the Standards Authority for Frontier AI, or SAFA. Its expected role would be to develop safety standards for advanced AI models, test them before release and establish procedures for responding to dangerous incidents. If the project goes ahead, it could become one of the technology industry’s most serious attempts to define its own limits. At the same time, it would create a major experiment. Can the AI industry really regulate itself?
+++ Read Dario Amodei's statement and plan on slowing down AI here +++
How the idea began
The companies have been negotiating for several months following a July 2026 proposal by Google DeepMind head Demis Hassabis. He suggested creating an independent standards body inspired partly by the US Financial Industry Regulatory Authority, or FINRA.
AI is not the stock market, but the principle is similar. Companies in a rapidly developing industry need common rules to make their work safer and more transparent.
OpenAI has confirmed talks with Google and Anthropic. SAFA could develop safety standards, test models before release and establish procedures for reporting dangerous incidents. For now it remains only an idea, with its leadership, funding and authority still unclear.
Why are competitors joining forces?
Google, OpenAI and Anthropic compete for investment, talent, computing power and users. Frontier AI, however, is different from an ordinary technology race. If one firm makes a serious mistake, the consequences could affect the wider digital infrastructure, not just its own users.
Against this background, technology leaders are calling for a slower pace of development. Anthropic CEO Dario Amodei urged the industry to slow the development of its most powerful models. His position was supported by Google DeepMind chief Demis Hassabis and OpenAI CEO Sam Altman.
I agree with Dario that we need to pace the frontier.
Sam Altman
He later said that slowing down had become a major topic of discussion within OpenAI. This creates a paradox. The companies profit from rapid development, but they also acknowledge that excessive speed could make AI more dangerous.
However, not everyone supports international or centralised control. During the United Nations General Assembly on September 22, 2026, US President Donald Trump rejected attempts to create a global system for controlling AI.
The United States totally rejects any attempt to construct a globalist scheme to control artificial intelligence.
Donald Trump
"Super Intelligence"?
Trump also said that the United States would not hold back a technology that could become greater than the Industrial Revolution. He called for continued technological competition and announced that official US documents should refer to AI as “super intelligence.”
This contrasts with Amodei’s and Altman’s calls for caution. While technology leaders discuss common safety standards, the US administration is emphasising national leadership and competition with China. SAFA is thus developing in a political atmosphere where many recognise the need for control, but few agree on who should establish the rules.
Self-regulation or a new cartel?
Companies outside the negotiations fear that Google, OpenAI and Anthropic could set rules for the entire industry. Smaller start ups, research centres and open source developers might then have to follow standards they had no role in creating. Critics are already asking whether SAFA could become a closed club for the market’s largest players.
Conflicts of interest are another concern. If the organisation is funded by the companies it oversees, how independent could it be? Would it criticise its members publicly? Unlike a government regulator, SAFA would probably lack the power to issue fines or ban dangerous systems. It could become merely a platform for voluntary recommendations, which are very different from binding oversight.
A final chance to build trust
SAFA could be useful only if it is transparent, independent and open to governments, universities, civil society, smaller companies and open source developers. It would also need clear evaluation criteria, public safety reports and independent experts..
For now, SAFA is only a proposal. Yet the negotiations show that Google, OpenAI and Anthropic understand that unlimited competition is becoming increasingly risky. The central question is whether they will accept rules that could limit their own commercial interests. Creating an organisation is easy. Giving it enough authority to control its creators will be much harder.
Featured photo by Jernej Furman / CC BY 2.0 — https://creativecommons.org/licenses/by/2.0/
0 Kommentare