The true meaning of sovereign AI power

AI governance is becoming a contest over access. Building the strongest systems confers power.

Power in the age of artificial intelligence (AI) is being built in chip factories, research labs, data centres and export control lists. The question is no longer whether AI will change the world. It is whether the rules for that world will be written by California’s AI firms, Beijing’s ministries and regulators in Brussels, or by a wider community that includes countries like India.

AI governance will emerge through repeated bargains over access, safety and public use. [Photo credit: Bloomberg]

AI governance is becoming a contest over access. Building the strongest systems confers power. Increasingly, so does the ability to decide who may use them and on what terms.

On July 6 and 7, the United Nations convened its first global dialogue on AI governance in Geneva. Its value was to give every government a place in a debate that markets and great powers have so far dominated.

Washington speaks of safety while seeking to preserve its lead. Its action plan stresses innovation, infrastructure and exports to allies. Recent US decisions on access to advanced Anthropic models showed that these systems are not ordinary software. A model available one day can be restricted the next. Access is becoming an instrument of power.

Beijing has reached a similar conclusion, while presenting AI as a global public good. On July 16, 29 countries signed an agreement to establish the World Artificial Intelligence Cooperation Organization, or WAICO — a Shanghai-based intergovernmental body focused on global AI governance. India was not among the signatories. WAICO gives Beijing both a means to shape the rules and an institutional counterweight to Pax Silica, the US-led technology supply-chain initiative India joined in February.

On July 17, Xi Jinping outlined China’s global AI vision, urging cooperation and open-source development while promising developing countries AI training opportunities. He warned that unequal access could create a “new historical injustice” and opposed invoking national security too broadly in AI. That principle may be tested by Beijing itself. Chinese authorities have reportedly discussed restricting overseas access to some of China’s most advanced models, although no decision has been announced.

The UN’s independent scientific panel estimates that the US accounts for 75% of the computing power among the world’s top 500 AI supercomputers. China accounts for another 15%. The rest of the world shares the remaining 10%. Most countries are debating rules for capabilities they do not control.

Europe’s leverage comes from its market and laws. The EU’s AI Act may travel farther than European models. Brussels can influence AI use without controlling much of the technological frontier.

Calls for AI arms control often borrow from the nuclear experience, but the analogy has limits. Nuclear material can be counted and missile silos watched. Models can be copied, distilled, concealed or accessed through the cloud. The most dangerous uses may be invisible: a cyberattack or help in designing a pathogen. The checks used in nuclear arms control cannot simply be carried over to AI. Safeguards are possible but will have to work differently.

Common safety standards are necessary. Who writes them? If the countries controlling the strongest models dominate the process, access to their markets and technology could depend on standards they write. WAICO’s structure is unclear. It gives developing countries seats. A seat is not the same as influence. The test is whether members can shape its priorities or are expected to endorse an agenda set in Beijing.

India need not choose the same partner for every AI question. American research and advanced models matter. Europe has experience in rights and redress. Xi’s plans for AI application centres with BRICS and the Shanghai Cooperation Organizatipn (SCO) bring Beijing’s agenda into forums that include India. Engagement, however, must not create dependence on Chinese platforms in critical infrastructure or sensitive public systems.

Sovereign AI is often invoked loosely. It cannot mean producing every chip, model and application at home. For India, it means retaining meaningful choices while keeping public authorities answerable to Indian law. Sensitive public systems belong within Indian jurisdiction and require independent oversight.

Models used by government require testing in Indian languages and audits for bias and error. If AI screens welfare claims, citizens have a right to know who built the system, what data it was trained on and how it shaped the decision. They are entitled to an explanation, human review and an appeal in their own language. Where rights or benefits are at stake, a public official must remain answerable for the final decision. In a democracy, sovereignty becomes real only when the state remains answerable to the citizen.

India’s credibility in global AI governance will depend on how it governs AI at home. It has expanded subsidised compute through the IndiaAI Mission, started work on an AI Safety Institute and issued governance guidelines. Enforceable rules for high-risk public uses still lag behind. An advisory tool for farmers does not raise the same questions as a system affecting welfare eligibility or judicial decisions. The rules need to distinguish between them before such systems become entrenched.

WAICO’s arrival adds urgency to India’s diplomatic response. Building on its AI Impact Summit declaration, India can seek subsidised compute access, joint safety testing and safeguards for AI in multilingual public services across the Global South.

AI governance will emerge through repeated bargains over access, safety and public use. India should enter them with capabilities and proposals of its own. Otherwise, the age of AI may arrive with Indian users and Indian data, but without an Indian voice.

Syed Akbaruddin served as India’s permanent representative to the United Nations in New York and is currently dean, Kautilya School of Public Policy. The views expressed are personal