Sam Altman has reportedly told OpenAI staff that the current approach—testing AI models, releasing them, and flagging concerns—is only pacing 1.0. As AI grows more powerful, he says the world may have about a year to choose a stronger safety model. The core message is that AI safety *cannot remain optional* or limited to one country.
Altman reportedly described three possible futures. One is a global watchdog, similar to the International Atomic Energy Agency, to inspect AI systems, require audits, and monitor computing power. Another is a merger or partnership of leading AI labs under a CERN-like international organisation. The third is a dangerous race where one company becomes so dominant that it can effectively shut out all rivals—Altman called this outcome “terrible.”
A global watchdog appears to be the most practical option because it can preserve competition while setting common safety rules. But it will require cooperation among major AI powers, including the US and China, and independent powers to verify companies’ claims. A merger may reduce risky competition, but it could also place excessive control* over AI in *too few hands.
The worst outcome would be one private lab gaining a decisive lead and controlling the chips and computing power that others need. That would not be regulation; it would be monopoly power over a technology shaping jobs, security, elections, and daily life. Anthropic’s Jack Clark has also warned that the window for safeguards may be only “a small number of years.”
AI SAFETY CANNOT BE LEFT TO A RACE—IT NEEDS GLOBAL RULES BEFORE THE RACE DECIDES FOR US.
Sanjay Sahay
Have a nice evening.