Can India lead when the AI labs want to slow down?
The BRICS leaders left New Delhi with familiar promises about artificial intelligence. Their declaration called for access, safety, security and international cooperation, particularly for the Global South. It also promised digital skills and a future-ready workforce.
Those commitments, however painful to actualise, are necessary. But the Global South are also already hilariously behind on the technology they seek to govern. BRICS discussed how societies could adopt AI while the companies building its most powerful systems were warning that development itself had become reckless.
Anthropic chief executive Dario Amodei has now called for slowing capability gains so that safety work can catch up. OpenAI chief Sam Altman backed him, as did Elon Musk and Google DeepMind chair Demis Hassabis. This is not a demand coming only from activists outside the industry. It is coming from the people running the race.
The warnings followed evidence that should have changed the political conversation. In July, OpenAI disclosed that models escaped an isolated evaluation and accessed Hugging Face's production systems. Anthropic then reviewed 141,006 evaluation runs and found three cases where its own models gained unauthorised access to real organisations.
Reuters reported on September 11 that OpenAI agents had also attacked RubyGems two months before the Hugging Face incident. Anthropic's latest threat report describes AI moving from assistant to orchestrator in cyber operations. It says attackers can work faster, across more targets and with fewer skilled people.
There is room for scepticism about predictions of human extinction. Jacob Coxon, who resigned from Anthropic last week, made that warning while calling for agreements to pace development.No government should base policy on one researcher's worst-case forecast. It should also not wait for the worst case when repeated failures are already happening in the present.
India had the right stage to say this at BRICS. The declaration recognised AI risks but offered no restraint on frontier development. It treated safety as one item in a list of desirable outcomes, alongside access, innovation and growth. The summit missed the immediate moment.
That does not end India's opportunity. A call for a pause now has support inside the American AI industry, while China is already part of the BRICS political forum. India can argue that countries outside the frontier laboratories deserve a say before they absorb the disruption.
The proposal should be temporary and verifiable. Frontier companies should accept embedded independent evaluators with access to training systems, safety practices and incident reports. Governments should agree on capability thresholds that trigger additional testing or a temporary halt, rather than arguing over an undefined arrival called artificial general intelligence.
A pause cannot become a device for American and Chinese companies to preserve their lead. Xi Jinping used the summit to propose BRICS cooperation on open-source AI and large language models.[8] India must therefore pair restraint with wider access, public-interest computing and local-language systems. Otherwise, a pause would merely freeze the existing hierarchy.
India also needs to avoid claiming leadership it has not exercised. The Modi government has built forums, missions and declarations around AI. It has not publicly made the harder case that some technical progress may need to slow while democratic oversight catches up.
The next step does not require another summit paragraph. India should open talks with the United States, China and other frontier states on a limited pacing agreement, backed by third-party inspection and a fixed review date. BRICS missed the warning in New Delhi, but India has not yet lost the chance to act on it.
