Leaders of major artificial intelligence companies urged governments to establish international safeguards during a United Nations Security Council meeting in New York, warning that increasingly capable systems could cause widespread harm if they are developed or deployed without effective oversight.
OpenAI chief executive Sam Altman and Anthropic chief executive Dario Amodei described both the benefits and the potentially severe risks of advanced AI. They argued that governments cannot leave safety decisions entirely to the companies building the technology and should prevent control from becoming concentrated in a single firm or country.
A warning paired with a promise
The executives presented AI as a tool that could accelerate scientific discovery, expand access to expertise, and give people new creative and economic opportunities. At the same time, they raised the possibility that future systems could become difficult to control, be misused at scale, or give a small group outsized influence over essential decisions.
Those paired messages reflect the central policy challenge. Governments want the gains associated with a fast-growing industry but face uncertainty about how powerful the next generation of models will be. Rules written too narrowly may become obsolete, while rules written too broadly could protect established companies, slow useful research, or push development into less transparent settings.
Governments remain divided
British Foreign Secretary Ed Miliband said the United Kingdom plans to place AI safety and a common set of global standards at the center of its G20 leadership next year. His argument is that systems capable of crossing borders require at least some shared expectations for testing, incident reporting, and human responsibility.
The United States opposed creating a new global governance structure. White House science adviser Michael Kratsios said concern about AI should not become a reason to pause development or impose new international constraints. China’s representative emphasized national choice and open-source systems while warning against blocs that could force countries to align with one technological camp.
What workable safeguards might include
International cooperation does not require every country to adopt identical laws. A practical starting point could include common methods for evaluating dangerous capabilities, channels for reporting serious incidents, protections for critical infrastructure, and agreements that retain meaningful human control over decisions involving weapons or other life-and-death consequences.
Verification will be difficult. Unlike nuclear materials or large industrial facilities, advanced software can be copied, modified, and distributed rapidly. Governments also have strategic incentives to conceal capabilities. Any durable system will need independent technical expertise, carefully defined thresholds, and consequences that do not depend solely on voluntary company promises.
Developing nations have an additional concern: safety rules should not become a barrier that reserves advanced tools for wealthy countries. Access to computing, training, local-language systems, and meaningful participation in standards will influence whether governance is seen as legitimate.
The Security Council meeting did not produce a binding framework, but it clarified the competing positions. Industry leaders are publicly asking governments to share responsibility for the technology they created. Governments agree that AI will alter economic and national security, yet they remain divided over who should set the rules. The next phase will test whether broad warnings can be translated into specific obligations before the most capable systems are widely deployed.