In March I argued that in AI, geopolitics follows architecture. Own one layer of the stack, and you own everyone building above it. On Friday, that stopped being theory. A US export-control directive pulled Fable 5, the most capable model Anthropic had ever released to the public, away from every foreign national on Earth. Inside America or outside it. Even Anthropic's own non-citizen staff. Three days after launch, at 5:21pm, the model layer became a border. I'm an Indian founder building in Europe. That border runs straight through me, and through most of the world's researchers and builders. This is the entire case for a third force in AI. India and Europe will never win by outspending Washington or Beijing. We win by out-connecting them: sovereign models, European compute, shared infrastructure no single government can switch off after close of business on a Friday. You cannot build a knowledge economy on a layer someone else can revoke. I wrote the full argument three months before any of this happened: mycel-ai.de/en/blog/whitepap…

AI labs in India and Europe needs to stop thinking how to build better open source or more culturally rooted models and think really forward into building most capable model in the world unapologetically frontier... Then whether you open source it or use with sovereignty can also happen. India and Europe needs to build their own Mythos level or beyond capable models. Learn from China, I guess they knew it better has built so many amazing labs and models and open sourced it so we have something to fall back on for the time being. Even then they are 6 to 12 months behind Mythos... Time to think BIG.

India desperately needs an LLM of its own. But then, who's building one to compete with Claude?

If fable 5 is too intelligent to allow global access... Does that mean we have hit a model capabilities limit for equitable access - GPT 5.5, Opus 4.8, Gemini 3.5 Pro.... Thats it? What if GPT 5.6 or 6 is as good as Fable 5? What about Gemini 4.0 Pro?

Well this is massive: - Every non-US company sees how they can be cut off from US vendors with a snap of a finger. Non-US vendors getting free advertisement - What does this mean for US AI labs’ offices and employees outside the US? Reads pretty draconian What a mess

This is the template, not just the analogy. America denied us the cryogenic engine in 1993. We flew our own in 2014 and became a launch power that now lifts the world's satellites. Twenty years, one denial, total reversal. Sovereignty was never about making everything ourselves. It's about owning the one layer that can be switched off on us. In 1993 that layer was a rocket engine. Today it's a frontier model. Fable 5 going dark for every Indian on a Friday isn't the setback. It's the starting gun. And unlike 1993, we already have the playbook: UPI, RuPay, BharatNet, the fastest 5G rollout on earth. Public rails, built at home, that the world now copies. Build the model the same way.

❌ ❌ ❌ ❌ ❌ ❌ ❌ ❌ ❌ ❌ ❌ ❌

"The GOI isn't in the business of building advanced IT" is the part that doesn't hold. It built UPI. It built RuPay, now taking real share off Visa and Mastercard. It laid fibre to 2.15 lakh villages and ran the fastest 5G rollout on earth: 99.9% of districts, data at ₹8 a GB. None of that was the private sector alone. It was public rails, private innovation on top. That's not government vs private. It's India's actual model, and it beat duopolies the US never even challenged. Sovereign AI is the same shape: the state sets the rails, Sarvam and others build on top. India isn't the US. Our edge is that we have this playbook. Use it.

On 12 June, two of the world's most advanced AI models were switched off for the entire planet. Not by accident. By order. A government issued an export-control directive barring any foreign national, inside or outside its borders, from using them. With no way to verify nationality across a shared service in real time, the provider had no choice but to switch the models off for every customer worldwide, Americans included. The frontier of machine reasoning, the same engines now accelerating medicine, materials, and climate science, made unavailable to 8 billion people with a few hours' notice. This particular block may be lifted within days. The precedent is permanent: a frontier model can now be removed from the world by a single government's letter. For years we treated access to the best tools as a settled fact, like electricity in the wall. It never was. Open science, equitable discovery, the idea that a researcher in Nairobi or Naples should reach the same frontier as one in California, cannot rest on infrastructure that one capital can switch off. Here's what gives me hope. Europe is not short on talent. Not short on science. Not short on values, the public-interest, rights-respecting model the world actually wants to build on. We have been short on one thing: the urgency to fund our own frontier. That excuse just ended. So this is my open call. To investors: European frontier labs and sovereign compute are not a moral nice-to-have anymore. They are critical infrastructure. Back them like the grids and bridges they are. To policymakers: move at the speed of this moment. Compute, capital, talent visas, procurement that chooses European builders. A model served from Europe cannot be switched off from abroad. To founders and researchers: build. Ambitiously, urgently, now. The world is suddenly looking for an alternative. That window will not stay open forever. I build research infrastructure in Europe because I believe discovery belongs to everyone, not to whoever owns the servers. 12 June made the case better than I ever could. Let's not waste it. #SovereignAI #OpenScience #EuropeanDeepTech #AIpolicy

Every great discovery starts with a question, but often, it gets stuck at "how do I fund this?" That's why we built ResearchArk. Find funding. Write good applications. Build your network - all in one place. Join our intro webinar to see how it works: beta.researchark.eu/events/f…

Claude Code and Codex are great tools and the subscription based system with 5hr and weekly limit system is very economic way to use it. However, the limits are less transparen obviously due to a hidden variable subsidy. But i believe the system also favors over use, for example people may simply default to topmodel with maximum thinking budget. What is actually needed is a new kind of auto mode (may be premium auto and economy auto), that combines model selection, subagents, thinking budgets all in one so users dont have to default to most expensive modelwith max thinking for all silly things. @AnthropicAI @OpenAI

INDIA is ADANI, ADANI is INDIA. 😇

After 1 day of extensive testing... Only troubled by the usage limits in max.. Fable 5 is a beast.. Biggest noticeable change in coding ever, closest would be sonnet 3.5 to 3.6 jump, that was a minor update but it had made real enterprise quality coding possible. Opus 4.5 made it possible to just never needing to use a IDE or look at code. You could just trust it. Then gpt 5.5 came on to same level and even took lead over opus 4.8 even. Codex goal made days long tasks possible (not just coding).. But now fable 5 is taking all that together and taking it to a higher level, its about quality, precision and speed. Its able to trace well hidden performance bottlenecks and solve it.. 1000s times performance gains in minutes of coding. Often with 1 line of code change. All previous models been looking at same codebase for months and caught nothing. Now fable is eating them for lunch. Usage limits are painful. Pricing is very high. But if there is enough paybacks to what you are building, fable can do a lot of magic.

Anthropic's long "fabled" Mythos 5 got genomics tools and about a week, mostly on its own. It assembled single-cell data for millions of cells across 138 species, trained its own machine learning model that beat one published in Science. It was one hundred times smaller. Read that again. Largely autonomous. One week. Smaller and better than the published state of the art. For most of the history of science, the binding constraint was us. Human hands, human hours, the years to learn a technique, the months to run the experiment. We built our entire research economy around that scarcity: the grant cycle, the slow accretion of papers. I spent years inside that machine, moving hundreds of millions of euros across borders, and I can tell you it assumes labour is the bottleneck. That assumption is quietly dissolving. The same family, Anthropic says, compressed a two-month code migration into one day for Stripe, produced drug-design candidates for nine of fourteen protein targets with no human guiding the work, and generated molecular biology hypotheses its own scientists preferred to the human alternatives four times out of five. One was later confirmed, independently, by a lab working on the same protein. I want to be careful, because wonder makes a poor analyst. These are Anthropic's own results, and the world is more stubborn than a launch post. But even discounted heavily, something real has moved. You can read that movement most clearly in what they chose not to release. The strongest version, Mythos 5 with its safeguards lifted, goes only to vetted cyber defenders and, for now, the US government. The model the rest of us can use quietly reroutes questions on cybersecurity and biology to a weaker one. The reason is plain: the same intelligence that designs a gene therapy can design a virus. We have reached the point where access to capability is itself a question of governance. Not a hypothetical for a future ethics panel. A product decision, shipped today. So here is what I keep turning over. If labour is no longer the scarce thing in science, then judgment becomes everything. Knowing which question matters, which experiment is worth running, which door to leave closed. The bottleneck shifts from the bench to the mind, and from the individual to the institution. Can our funding systems and our creaking research culture metabolise tools that no longer wait for us? I build in this space, so I am not a neutral observer. But I did not get into it because I love software. I got into it because somewhere tonight a researcher is staring at a deadline with an idea that might change something, and for the first time in human history they may have a collaborator that never sleeps. What we do with that is still ours to decide. It always was. That is the frabjous, terrifying gift of this moment: the tools are finally ready. The question is whether we are. ResearchArk researchark.eu anthropic.com/news/claude-fa…

It’s the question every researcher asks themselves at some point. As you look to streamline your workflow and advance your #research #application, explore our open beta and see how ResearchArk can help you move faster - from idea to impact! beta.researchark.eu/register… #publicbeta

For decades, we’ve explored Norse realms, Greek pantheons, and Samurai islands. But the global gaming industry is simply not ready for the sheer scale of India’s upcoming AAA revolution. 🇮🇳🎮 From the grounded, gritty vengeance of the Chola Dynasty in Son of Thanjai ⚔️: 🔗 youtu.be/V3HKrx1HZic?si=CjZl… To the cosmic, Unreal Engine 5-powered mythological warfare of Raji: Kaliyuga ✨: 🔗 youtu.be/V5TMzD6Ngnk?si=8SMZ… Homegrown studios are finally reclaiming the subcontinent's ancient epics. I just wrote an in-depth breakdown of the trailers shaking up the industry right now—including the photorealistic Unleash the Avatar. The land of stories has finally entered the chat. 🔥

Classic BJP IT-cell toolkit drop. Exact same “If BJP ruled 70 years instead of Congress, India would be 100 years ahead of China” post is suddenly everywhere in the last 72 hours. Reality check: • Congress PMs ruled 54.45 years • BJP/NDA ruled 18.25 years (Modi = 12 years) • The “70 years Congress” line is deliberate exaggeration. A fantasy “what-if” doesn’t erase 12 years of record. Stop selling imaginary history to dodge real scrutiny. #FactsOverFiction

Dear Anupam Kher, here are India's unique achievements under Modi: 1. Press freedom: 157th of 180 in 2026, down from ~140th in 2014 (RSF). 2. Democracy: downgraded to "electoral autocracy" (V-Dem) and "Partly Free" (Freedom House, 66/100 vs 77/100 in 2017). 3. Internet shutdowns: world's #1 for six straight years, 116 in 2023; 771 since 2016 (Access Now). 4. Global Hunger Index 2024: 105th of 127, "serious," behind Bangladesh, Nepal and Sri Lanka. 5. Rupee: ~59/USD in 2014 to ~95-96 in 2026, a fresh record low. 6. Demonetisation (2016): 86% of cash voided; ~99.3% returned per RBI, killing the black-money rationale. 7. Electoral bonds: struck down as unconstitutional by the Supreme Court (2024); Rs 16,518 cr routed through it, BJP the top beneficiary. 8. ED probes: 115 of 121 opposition leaders targeted since 2014; 23 of 25 who joined the BJP saw cases dropped (Freedom House 2025). 9. COVID second wave (2021): WHO estimated ~4.7 million deaths vs an official 480,000. 10. Farm laws (2020-21): a year of protests, hundreds dead, all three repealed in a rare U-turn. 11. Manipur (since 2023): 200 dead, a months-long internet blackout, the PM criticised for his silence. 12. Passport: still 75th globally, visa-free access to just 56 countries. 13. Thailand: scrapped visa-free entry for Indians in 2026. A unique identity on the global stage, indeed. Sources: RSF, V-Dem, Freedom House, Access Now, Global Hunger Index, RBI, Supreme Court of India, WHO. PS: Yes, yes, every one of these indices is secretly funded by George Soros. The RBI, the Supreme Court and the WHO too, apparently. Powerful man.

We built ResearchArk to connect you with researchers across the globe - opening doors to collaboration, mentorship, and new opportunities. Looking for partners in your application? Check us out, we are in public beta as of June! beta.researchark.eu/register… #Research #PublicBeta