Frontline ·
AI’s atom bomb moment?
The AI revolution raises questions about ownership, labour and access as a few firms control technology built on collective human knowledge.
Late in 1944, a Polish physicist named Joseph Rotblat packed his bags and left Los Alamos, the Manhattan Project’s laboratory and the birthplace of the atomic bomb. He had joined the project to build the atomic bomb because, like many scientists then, he feared Hitler would build one first. When it became clear that Germany was nowhere close, he saw no reason to stay. He was the only scientist known to have left the Manhattan Project on ethical grounds. The bomb was built anyway, and it fell on Hiroshima the following August. Rotblat spent the rest of his life speaking up against the bomb. He was the youngest signatory of the 1955 manifesto in which Bertrand Russell and Albert Einstein asked scientists to “remember your humanity, and forget the rest”, and in 1995 he shared the Nobel Peace Prize with the Pugwash Conference. Rotblat was 36 when he left the Manhattan Project.
Last week, 27-year-old Jacob Coxon, a British researcher, walked out of Anthropic, the AI giant that develops the Claude family of large language models (LLMs). Coxon spent three years training LLMs, first at OpenAI and then at Anthropic, and on September 8, he wrote on X that neither company was acting responsibly; they were, he said, “racing straight to self-improving superintelligence and gambling with our lives”.
Evan Hubinger, an alignment science lead at Anthropic, responded to Coxon’s concerns saying he would put the odds of AI killing everyone within a decade at more than 10 per cent. On September 12, Anthropic’s chief executive, Dario Amodei, published an essay asking the industry to “pace the frontier”, and the heads of OpenAI, Google DeepMind, and xAI said they agreed.
The (mostly) men who build these machines say the machines frighten them. Their remedy worries me more. OpenAI’s chief executive, Sam Altman, has hinted at a pact among the leading labs to jointly slow down. If that happens, a few companies, nearly all of them American, will decide among themselves how fast a technology moves that will directly touch, at last count, more than eight billion lives, most of which have never typed a prompt. Nobody plans to ask the paddy farmer in Palakkad or the delivery rider in Nairobi.
In the 1980s, trade unions in Kerala, many of them led by the Left parties, marched against the computerisation of banks and public offices. They feared the machines would swallow up jobs, especially of the poor and the vulnerable. And some jobs did go. But by 2002 the State had launched the Akshaya e-governance project in Malappuram, with the aim of making at least one person in every family IT-literate. To be clear, the unions were not all wrong about capitalism, but their mistake was to believe a technology could be halted by standing in front of it. It can’t. We can only fight over who owns it and whom it serves, and that fight takes patience.
The race Coxon is worried about is the one for returns. Anthropic is preparing for a stock market listing, with reports putting its target valuation at $2 trillion (comparable to the GDP of several countries, including Australia, Mexico, Spain, and South Korea). Anthropic has also signed deals for up to 5 GW of new computing capacity each, with Amazon, Google, and Broadcom. When Anthropic is obliged to answer to investors every quarter, it will struggle to slow down for long, regardless of what its founders feel on a Saturday.
This is why I feel that the future of AI is socialist. The word should not scare anyone in London or Palo Alto. It does not mean five-year plans for chatbots. It means only that a technology made from the common stock of human knowledge must answer to the common good, and that publics, not boards, should decide its limits. There is a branch of study called AI CI, short for collective intelligence or common intelligence.
In the first volume of Capital, Karl Marx introduced the idea of “primitive accumulation”, the enclosures and plunder that turned common land into private wealth before the factories came. The AI giants and their labs have done something similar with the written record of the human species. They have scraped books, research material, artworks, news, forums, and codes, mostly without asking. Last year Anthropic agreed to pay $1.5 billion to settle a suit by authors whose pirated books had gone into its training data. But with around half a million books mentioned, it worked out to about $3,000 a book. The labour comes cheap too. In 2023, TIME reported that Kenyan workers hired to label text as violent and abusive for OpenAI were paid less than $2 an hour. Kate Crawford, in her book Atlas of AI, calls the technology “neither artificial nor intelligent”. It is made of minerals, water, badly paid human labour, and other people’s words and drawings.
Out of this has come a new class divide. The world is now split into the prompt haves and the prompt have-nots. A report by Microsoft’s AI Economy Institute found that in the first three months of 2026, 27.5 per cent of working-age people in the Global North used generative AI against 15.4 per cent in the Global South. An earlier Microsoft report put the number of people worldwide who lack electricity, connectivity, and skills to use AI at nearly four billion. Marx had predicted that “general intellect”, the knowledge of society as a whole, would become a direct force of production. That much has come true. The knowledge in these machines came from all of us, but now a handful of firms hold the title deeds.
The data these firms use has no nationality, neither does AI. A model trained on Malayalam blogs can answer questions in Mombasa. So rules made in one country cannot govern it. It needs a universal, common approach. The United Nations has made a start in this regard.
In August 2025, the General Assembly set up an Independent International Scientific Panel on AI, 40 experts co-chaired by the Canadian computer scientist Yoshua Bengio and the journalist and human rights activist Maria Ressa, along with a Global Dialogue on AI Governance, which held its first annual session in Geneva this July and meets next in New York in May 2027. But the panel writes reports and holds dialogues. Neither can tell a lab to stop.
What we need will look more like the International Atomic Energy Agency set up in 1957: a standing collegium with inspectors and a mandate, able to set limits on the most dangerous systems and spread the gains from the rest. Its seats should not be weighted by market value. A fishworkers’ cooperative in Kerala and a nurses’ union in the Philippines have as much stake in these systems as an executive in San Francisco. Any slowdown that the labs agree on should be brought under such a body, where the rest of the world can see the terms and change them.
India has a fair claim to chair this conversation. If the government wants a viswaguru moment, this is one worth having. In February, India hosted the first global AI summit held in the Global South, and close to 90 countries and organisations signed a declaration calling for the “democratic diffusion” of AI and affordable access to its basic resources. Nehru’s India went to Bandung in 1955 (to the Asian-African Conference) to shape a post-colonial global order, the same year Russell and Einstein wrote their manifesto, and helped build a bloc of nations that the great powers preferred to ignore. But a moderator has to be trusted by all sides, and India’s entry into the US-led Pax Silica initiative will make some capitals in the South wonder where it actually stands.
Rotblat could not stop the bomb. But some things emerged—the partial test ban in 1963, the non-proliferation treaty in 1968, etc. The work was slow and painstaking, but AI will need that kind of patience. A table with more chairs around it is needed to make rules that follow the data across borders.
Meanwhile, we are still seeing some of the worst fallouts of AI-generated fake news. Read Rituparna Banerjee’s piece “Synthetic misogyny” on the sexualised campaign against Sonia Gandhi using AI images that were widely shared by official BJP handles.
Write back, as usual, with your comments and concerns.
Jinoy Jose P.