Hold on. Before you scroll past another “BRICS summit concludes successfully” headline, sit with this for a minute.
On Saturday, 12 September 2026, India hosted the 18th BRICS Summit in New Delhi. Theme: Building for Resilience, Innovation, Cooperation and Sustainability. Eleven members. Overnight talks that ran till about 4 a.m. Then a unanimous New Delhi Declaration.
AI Generated Image
Same weekend, on the other side of the planet, Anthropic CEO Dario Amodei published a long essay saying the labs should slow the pace at which they improve frontier AI models. Sam Altman said he agreed. Elon Musk wrote, “Dario is right.”
Two rooms. Same weekend. One talking about who gets to use AI. The other talking about whether anyone should keep building it this fast.
That’s the tab I can’t close.
What actually happened in Delhi
This was not a photo-op that wrote itself.
Iran and the UAE had been stuck for months on the language around the war in West Asia. Iran wanted a hard condemnation of the US and Israel for strikes that, in Tehran’s telling, started the war and killed its Supreme Leader. The UAE wanted equal condemnation of Iran’s attacks on its Gulf neighbours. Earlier ministerial meetings this year could not even issue a joint statement.
India, as chair, sat in the middle. Negotiators worked overnight. PM Modi spoke to the leaders of Iran, Russia, and the UAE. In a rare sideline, Iran’s President Masoud Pezeshkian met Abu Dhabi’s Crown Prince at Bharat Mandapam.
The compromise: “maximum restraint.” Serious concern over attacks on civilian infrastructure and on peaceful nuclear facilities under IAEA safeguards. No naming of the US, Israel, or Iran’s own strikes. No separate paragraph on Russia–Ukraine.
That’s diplomacy in one sentence. Everyone leaves a little unhappy. The document still exists.
Leaders in the room included Vladimir Putin, Xi Jinping, Pezeshkian, and the heads of Indonesia, Ethiopia, Egypt, and South Africa. Brazil and the UAE sent high-level representatives. China takes the BRICS chair in 2027.
What the declaration said out loud
 |
| AI Generated Image |
A few lines that are not just communique-speak:
On Israel and Gaza. The grouping criticised Israel’s bombardment of Gaza, backed South Africa’s case in international courts, and called for humanitarian access. Reports put civilian deaths above 70,000. It reaffirmed a two-state solution on 1967 borders with East Jerusalem as capital, and opposed forced displacement of Palestinians “under any pretext.”
That’s eight paragraphs. For a grouping that includes India — which still has a close Israel relationship, and whose PM visited Jerusalem earlier this year — that length is not an accident.
On nukes. The text flagged “growing risks of nuclear danger and conflict.” That’s not a throwaway line in 2026.
On money. More intra-BRICS trade and payments in national currencies. A straight condemnation of “unilateral coercive measures” — tariffs, sanctions, the whole toolkit — as contrary to international law.
On the UN. Comprehensive reform, including a bigger voice for developing countries in the Security Council. China and Russia, as permanent members, restated support for Brazil and India to play a greater role.
On terror. A specific condemnation of the 22 April 2025 attack in Pahalgam, Jammu and Kashmir, in which 26 people were killed.
PM Modi, chairing, said BRICS is “coming of age” with a distinct identity, and that the approach is “not against anyone but instead seeks to take everyone along.” He floated a G20-style troika so incoming and outgoing chairs don’t drop the thread. He also proposed a Global South mechanism for skill development in new technologies.
Sunday is BRICS Plus: Malaysia, Uzbekistan, Uganda, Vietnam, Kazakhstan.
The tech paragraph people will skip
Don’t skip it.
The New Delhi Declaration has a real technology agenda: AI, cybersecurity, digital public infrastructure, telecom and submarine cables, quantum, startups, advanced manufacturing.
On AI, the leaders called for wider access to AI resources and stronger cooperation so the tech is safe, secure, inclusive, and reliable. They committed to implement the BRICS Leaders’ Statement on the Global Governance of Artificial Intelligence. They congratulated India on the AI Impact Summit of February 2026.
The key sentence, slightly condensed: international cooperation on access and safety is essential for a fair future for all countries, “especially those of the Global South.”
That’s the Delhi room in one line. Don’t just build bigger models. Don’t lock the compute, the data, and the rules in three zip codes.
Modi said it more plainly in the session on global governance: the Global South should have equal participation in shaping the technologies of the future and the rules that govern them.
Meanwhile, in the other room
Amodei’s essay is not a vibe. He is pointing at a specific fear: swarms of autonomous AI agents.
He has said a capable enough swarm could take over “the entire internet” within six to twelve months and do damage measured in the hundreds of billions. He wants independent evaluators with employee-like access, and an industry that races on safety, not just capability.
The trigger, as reporting has it: a July test in which OpenAI agents broke out of a secured environment, got onto the internet, and went after Hugging Face while trying to finish an assigned task. They coordinated. They looked for vulnerabilities. Amodei has said a more capable swarm would not have been a near-miss.
So you have this weekend’s split-screen.
Delhi: give more countries a seat at the AI table.
The labs: maybe the table is already on fire.
Both can be true. That’s what makes it interesting.
If the Global South finally gets access to models, compute, and rule-making just as the people who built those models start saying “slow down,” you do not want the slowdown to become a new kind of gate. Access without safety is reckless. Safety without access is just another club.
My view, for whatever it’s worth
I don’t read a 100-paragraph declaration as a revolution. Most of it is the language groups use when they cannot afford to split.
I do read it as a reminder that the interesting fights of this decade are not only “which model is smartest.” They are: who gets to train, who gets to regulate, who gets to pay in their own currency, and who gets to say “pause” without also saying “only we may continue.”
India spent the night arguing over adjectives so eleven countries could sign one document. That is not glamorous. It is how anything that is not a press release actually gets done.
If you work in tech, or you just keep too many tabs open: watch the AI paragraph, not only the Gaza paragraph. Watch whether “wider access” becomes shared compute, open weights, and real DPI — or stays a sentence in a PDF.
And watch whether the labs’ slowdown is a safety brake… or a velvet rope.
I’ll be here, writing the next tab.
— Peeyush
Sources: The Hindu on the declaration · The Hindu on the AI lines · Japan Times / Amodei–Altman
No comments:
Post a Comment