Control over artificial intelligence is increasingly becoming a matter of international regulation. On September 29, U.S. President Donald Trump and the leaders of OpenAI, Anthropic, Google, Meta, xAI, and Nvidia signed an agreement on the safety of the most powerful AI systems.
The companies agreed to establish internal monitoring systems, test models for potentially dangerous capabilities, bring in independent auditors, and assign oversight to dedicated committees on their boards of directors. The main concerns include cyber threats, biological and chemical risks, and AI’s ability to interact independently with computer systems.
These safeguards are a response to the rise of autonomous AI agents. Tests have already produced cases in which models tried to gain access to computer systems or took actions their developers had not specifically instructed them to perform.
Although AI does not yet possess human consciousness or a human system of values, one question remains open: what happens if it one day decides on its own to hack into critical infrastructure or otherwise gain unauthorized access? It is probably not a question humanity wants answered through firsthand experience.
For now, the agreement is voluntary. It carries no penalties and provides no strict government oversight. But the law is always trying to catch up.
Across the world, AI safety is becoming more formalized, with new rules, institutions, and oversight mechanisms taking shape. The 2023 Bletchley Declaration laid the groundwork for international cooperation and joint assessments of the risks posed by advanced AI systems. The 2024 Seoul Declaration supported the creation and expansion of AI safety institutes, research programs, and oversight bodies.
That same year, 16 leading technology companies committed to assessing the risks of their models, setting thresholds for acceptable risk, and withholding models from release if those risks could not be brought below those thresholds.
Regulation is also reaching beyond the development of powerful AI systems into everyday use. Some countries already require labels for deepfakes and other artificially generated content.
Gradually, the rules are coming to cover the full range of AI applications, from oversight of the most powerful systems to the content people encounter every day. Let’s hope they don’t become too strict on that last point. We could lose a substantial share of the “slopaganda” featuring Trump as a superhero and his many other alter egos.
In “The Alternative,” limits on AI also become a subject of debate, but at a much more consequential level: legally binding cooperation between nations. AI is far more deeply embedded in critical infrastructure and has a much wider range of capabilities.
The AIs developed by different countries are also in constant contact, engaged in a rivalry that extends beyond the digital realm. It plays out in the physical world as well, affecting urban infrastructure, military systems, and robotic systems. That makes it a question both of controlling the technology and of finding a new balance among people, governments, and artificial intelligence.
In the book, “Yalta-2” becomes a turning point for humanity’s future. Russia, China, and the United States discuss not only spheres of influence, but also how to control, regulate, and constrain superintelligence across different areas of life. The novel calls its AIs “superintelligences”—the same term Trump has recently started using.
A similar process is beginning to take shape in direct talks between governments. In September, the United States and China agreed to establish a dedicated channel for discussing serious AI incidents and to begin a dialogue on superintelligence.
We are still a long way from “Yalta-2.” But international rules for technologies that, until recently, existed mostly in science fiction are gradually becoming part of the real political and technological agenda.