Media365
The last fortnight has exposed fault lines in the global debate on AI. At the UN Security Council, Anthropic’s Dario Amodei warned that, if managed poorly, AI could pose a risk to humanity as a whole, while OpenAI’s Sam Altman cautioned that humanity could lose control of its future. Rarely does an industry seek its own regulation. Yet, in the same chamber, Michael Kratsios, the US president’s science and technology advisor, rejected any centralised international control of AI.
Days later, President Donald Trump hosted technology chiefs at the White House and unveiled a voluntary accord. Companies pledged internal model monitoring, a team to verify it, outside auditors and independent board oversight. Trump called the pledges “morally binding”. Self-regulation is now American doctrine.
Paradoxically, Washington has simultaneously opened a channel to Beijing. After the Trump-Xi summit, both agreed on an AI incident communication mechanism, with a dedicated dialogue due in November. Middle powers have moved further. Twenty countries and the European Union, including Germany, South Africa, Canada, Australia, the UAE and Singapore, called for keeping AI under human control, possibly through a global oversight body. Tellingly, neither the US nor China signed; nor did India.
Four models now compete. The EU offers binding, risk-tiered legislation. America favours market-led voluntarism. China practises state-directed, sector-specific control. A loose international layer of OECD principles, UNESCO recommendations, summit declarations and a UN scientific panel supplies vocabulary without obligation. None suffices alone.
The history of cyber governance is instructive. The UN Group of Governmental Experts (GGE), first convened in 2004, spent a decade establishing that international law applies to cyberspace. Its 2015 report produced 11 voluntary norms of responsible state behaviour: States should not knowingly allow their territory to be used for wrongful ICT acts, should not attack critical infrastructure or emergency response teams, should cooperate on requests for assistance, and should responsibly report vulnerabilities. The General Assembly endorsed them.
After the 2016-17 GGE deadlocked over self-defence and international humanitarian law, the General Assembly in 2018 created two tracks: A Russian-sponsored Open-Ended Working Group (OEWG) open to all members and a US-backed GGE. Both reaffirmed the 11 norms in 2021. A permanent UN mechanism has since succeeded the OEWG, still non-binding.
Two lessons follow. First, consensus norms are achievable only when the principal adversaries sit at the table. Second, a decade-long, consensus-driven process cannot keep pace with a technology whose capabilities shift every few months.
What would suit AI is a layered architecture rather than a single treaty: Binding national law where frontier laboratories operate; internationally agreed capability thresholds that trigger pre-deployment testing; mandatory cross-border incident reporting; an IPCC-style scientific body to establish shared facts; and, for the most capable systems alone, an IAEA-type verification regime anchored in compute monitoring. The 20-nation proposal approximates this.
However, such a regime cannot succeed without China, the only country besides America with genuine frontier capability. Its open-weight models already power applications across Asia, Africa and Latin America, placing them beyond any regime that governs only Western developers. Beijing’s approach is equally consequential. Domestically, it regulates tightly through algorithm registries, generative AI rules and mandatory labelling of synthetic content. Internationally, it presents AI as a development right, champions UN-centred governance and has proposed a world AI cooperation organisation, an offer many developing countries find attractive. A regime excluding Beijing would become a Western club, inviting a parallel Chinese bloc of standards. The nascent US-China incident channel is the minimum foundation; it must eventually extend into multilateral confidence-building. The middle powers must build an architecture both giants can later join.
India’s prudent course is to join any open framework, while conditioning oversight on equitable access to compute and models. It should use the credibility of hosting the AI Impact Summit and serving on cyber GGEs to bridge frontier powers and the Global South. It must strengthen its AI Safety Institute so that Indian evaluators shape testing regimes. And it should seek a seat in any incident-reporting framework, since harms originating elsewhere will land in Indian markets.
The cyber precedent shows that norms without enforcement still matter: They build habits of consultation and a common language. AI may not grant the world a decade to learn. Rule-making must be swift and inclusive.
The writer is a defence and tech policy adviser and author of The Digital Decades: On 30 Years of the Internet in India
