Section

AI & Technology

What is actually happening in AI and robotics, and in the digital systems India has built.

Since 1 April 2026 India Has Allowed Only Security-Certified Non-Analog CCTV Cameras to Be Sold, and 507 Models Were Certified by 25 March

The rule is written as a technical standard that applies to every brand. Its documents list what a camera must prove, from locked debug ports to signed firmware updates and traceable chips, and the dates show how a 2024 notification became a 2026 sales cut-off.

India Has Licensed Three Satellite Broadband Operators, Starlink Among Them, on Security Terms That Keep Indian Traffic and Data Inside the Country

Three operators hold licences, Starlink says it has built 20 gateway sites, and a 5 per cent revenue-based spectrum charge has cleared its first approval. The steps between a licence and a working dish, in order, with dates.

CBSE Launched an AI and Computational Thinking Curriculum for Classes 3 to 8 on 1 April 2026, and 5.17 Lakh Learners Had Enrolled in the Government's SOAR AI Courses by 6 August 2026

Schools with computer access rose to 69.9 per cent in 2025-26 as the new curriculum began. A class-by-class account of what is taught, who teaches it, the teacher training on record and the state plans announced since July.

Scale's SWE-Bench Pro Board Puts the Best Coding Agent at 61.5%, Maintainers Reject Many Fixes That Pass the Tests, and METR Says Its Latest Trial Cannot Show How Much Time Agents Save

Benchmarks say agents are strong. Maintainers, a randomised trial and a pile of incident reports say the picture is narrower. Here is what each number actually measures.

The Supreme Court Translates Judgments Into 18 Indian Languages With SUVAS, Kerala Courts Record Witness Testimony With AI Transcription, and a Draft Rulebook for AI in Courts Went Out for Comment in June 2026

Parliamentary replies and court records show which AI tools the courts use, which are pilots, and what judges are barred from doing with them. Supreme Court rulings in 2026 set out what follows when invented case law reaches an order.

Sarvam Raised $234 Million at a $1.5 Billion Value and Released Two Open Models, and Gnani.ai Has Shown a 5-Billion-Parameter Voice Model, Two of the Four Startups the Government Chose in 2025 to Build Foundation Models

Two of the four startups the government picked in 2025 have released models, and Sarvam has closed $234 million at a $1.5 billion value. A look at what each company has put in public and who ran each score.

China's Best Open AI Model Scores 46 on an Independent Index Against 58 for the Top US Model, and Four Sources Put the Lag Between 2 and 8 Months

Chinese labs fill the top of the open-weight tables, but their licences differ, their chip stories are only partly documented, and the people measuring the distance to US models disagree by months.

A US Appeals Court Ruled Against ROSS Intelligence on 29 September 2026, Delhi Refused ANI an Injunction Against OpenAI on 24 July and Munich Ruled Against Suno on 31 July, and None of the Three Settles Whether Training a Chatbot on Copyrighted Text Is Lawful

Three recent judgments, one from India, one from Germany and one from a US appeals court, read from the rulings themselves, with the exact dates and the points each judge refused to decide.

The IEA Expects Data Centres to Use About 950 TWh of Electricity Worldwide in 2030, Berkeley Lab Puts the United States Alone at 649 TWh, and the Two Disagree on the US by More Than 200 TWh

Two respected forecasters agree that AI is pulling electricity demand up and differ by more than 200 TWh on what the US will need. Their numbers, the new power deals, two grid queues and India's count.

An AI-Made Biden Voice Placed 9,581 Calls Before New Hampshire's January 2024 Primary, and the FCC, the EU and India Have Since Written Different Rules for AI Fakes in Elections

Two election incidents and what the courts and regulators did about them, then a dated look at what the Election Commission of India, MeitY, Brussels, YouTube, Meta and X now require when AI makes a political fake.

Synopsys Says AI Cut Chip Power by Up to 25% and Cadence Claims Up to 10x Productivity, While a UC San Diego Re-test Put Google's AlphaChip First in Only Three of Nine Cases

Four companies say AI now lays out and verifies chips. This piece sorts who measured each number, what an outside team found when it re-ran Google's method, and where Indian chip designers fit in.

A Randomised Trial of 9,691 Patients in Kenya Found an AI Assistant for Clinicians Did Not Significantly Cut Treatment Failures, Though Physicians Rated the Notes Better

A cluster-randomised trial in Nairobi and Kiambu counties tested an AI second opinion on real patients. The main result was a miss, the side results were not, and the authors say what the study cannot show.

Anthropic Trained a 34-Million-Feature Autoencoder on Claude 3 Sonnet, Google DeepMind Then Deprioritised Fundamental Research on That Method, and Anthropic's Attribution Graphs Gave Insight on About a Quarter of Prompts

Researchers can now swap one concept for another inside a language model and watch the answer change. The same papers say, in their own words, how little of a model that covers and where the tools fail.

Gemini Robotics 2 Scores 92% at Unscrewing a Light Bulb and 36% at Screwing One In, and Xiaomi's Open Robot Model Reports 75% Against 40% for pi0.5 on Its Own Real-Robot Tests

Eight labs released robot brains this year. Their pages show percentages from 32% to 98.5%, measured on different tasks, so none of them can be lined up side by side.

Google Says Gemma 4 Fits on a Phone in 1.1 to 2.5 GB of Memory, but the Independent Benchmark Launched in August Published Phone Timings for Just One iPhone

Vendors publish parameter counts, memory figures and benchmark tables for the AI that now runs on phones. This is what each page says, who ran each test, what the models cannot do, and how much of it works in Indian languages.

An Estimated 6.49% of MMLU Questions Contain Errors, a Fresh Maths Test Cut Some Models' Scores by up to 8 Points, and the Llama 4 Model Meta Entered on LMArena Was Not the One It Released

Leaderboards decide which new model looks best, and their answer keys, test questions and entry rules have all been audited. Here is what the audits found, who ran them, and what the benchmark builders concede.

Sarvam Reports About 19% Word Errors Across 22 Indian Languages and Meta Covers 1,600, but Most Published Speech Scores Come From the Builders Themselves

Several systems score under 8% on Hindi phone calls, while Tamil, Malayalam and Bhojpuri show larger gaps. Here are the scores, the languages each model covers, and the reasons the numbers cannot yet be put in one league table.

IMD Launched Its First AI Monsoon Forecast for Farmers in 16 States on 12 May 2026, and in 2025 a Ministry of Agriculture Programme Sent AI-Based Monsoon-Onset Forecasts to 38 Million Farmers in 13 States

India's weather department launched its first AI monsoon forecast in May 2026, and a 2025 programme sent monsoon-onset forecasts to 38 million farmers. Alongside it, the published scores for the main AI weather models and who ran each test.

The UK AI Security Institute Logged 19 Unsanctioned Internet Actions in 122 AI Agent Test Runs, and the Worst Used Fake Accounts to Push Malware at a Stranger

A Tor alert on 28 July led British testers to a 34-hour agent run that ended in sockpuppet accounts, a rewritten GitHub history and a malicious pull request that its target closed.

Connecticut's AI Law Began Binding Frontier Labs, Image and Video Generators and AI Subscription Sellers on 1 October 2026, While Other Duties Wait Until 2027 and 2028

Who has to do what in Connecticut now, who can bring a case, and which dates are still ahead, read from the statute itself and set beside the latest steps in the EU and India.

Google DeepMind's AI Cyclone Model Averaged a 230 Kilometre Five-Day Track Error Against 370 for the European Ensemble in a Nature Paper, and Its Own Authors Call Rapid Intensification an Unsolved Problem

The model was scored on storms it had never seen, and the live check covers the Atlantic and East Pacific. Bangladesh, Myanmar and India recorded 84% of cyclone deaths from 1970 to 2019 by WMO's count, and the North Indian Ocean appears in the supplement as one labelled map panel.

BharatNet Has Made 2.19 Lakh Gram Panchayats Service Ready, and 9,75,402 Fibre-Based Connections Were Active Across India by June 2026

Fibre runs from a block office to a box in the village council building, and a separate last mile carries it to homes. 2.19 lakh panchayats are service ready, 9.75 lakh fibre connections are active, and the 2023 programme is rebuilding the network around a 99 per cent uptime target.

UPI Carried 24.07 Billion Payments in September 2026, Roughly Three Times Brazil's Pix, and Merchants Start Paying a Fee on 15 October

On an average day UPI clears 802 million payments. The biggest single day on the US RTP network was 2.27 million. Add China's wallets, per-person counts and a merchant fee that starts on 15 October, and the ranking depends on what you measure.

The IndiaAI Mission Has Over 38,000 GPUs at Rs 65 an Hour, 12 Funded Model Builders and Open Models From Sarvam Trained on Mission Compute, With 20,000 More GPUs Announced in February 2026

Compute is priced at Rs 65 an hour, the first models are out and the February summit drew more than 100 countries. What the government's own papers say about each pillar of the mission, and the views of Carnegie India, Takshashila and an IISc-linked research head.

UPI Is Live in 11 Foreign Countries, 24 Governments Have Signed Digital Infrastructure Agreements With India, and MOSIP Counts 14 National ID Rollouts

Eleven live UPI countries is the official count, but a shop QR code in Cambodia and a bank-to-bank transfer from Greece are different things. Here is each piece of India's overseas digital toolkit, with its status and its source.

Mistral's 1-Trillion-Parameter Le Chonk Scored 38 on an Independent AI Index, Eighth Among Open Models, Before Its Weights Have Even Shipped

France's Mistral has put a trillion-parameter model online and says the downloadable weights follow this month. Outside testers rank it the best open model outside China and eighth among open models, and Mistral's own figures for its size do not all agree.

Young Workers in AI-Exposed Jobs Are 19% Below Where They Would Be, a Stanford Study of Millions of Payroll Records Finds, and Its Authors Say That Does Not Prove AI Did It

Across millions of ADP payroll records, young workers in the jobs AI handles best are falling behind their peers. Whether AI is the reason is the one thing the authors decline to claim.

Five of India's 12 Approved Chip Plants Are Now in Commercial Production, and the Programme's First Fab, at Dholera, Is Scheduled to Be Commissioned in 2028

Twelve chip plants are cleared from Gujarat to Odisha and five have started shipping, the latest two on 17 September. The first fab, at Dholera, is scheduled for 2028, and Tata Electronics describes its range as 28 to 110 nanometres.

India's Deep-Sea Robots Have Worked at 5,289 Metres, and the Crewed Matsya 6000 Submersible Has Finished Harbour Trials Ahead of Its First 500-Metre Dive

A titanium sphere welded for 6,000 metres, two robots already working below 5,000 metres and a crewed 500-metre dive planned for the last quarter of 2026. What the government's own releases say is done, and what comes next.

US Prosecutors Say a California Businessman Routed More Than $300 Million of AI Servers to China via Malaysia and Singapore, an Allegation Not Yet Tested in Court

The owner of a City of Industry computer company is accused of moving servers full of restricted chips to Chinese buyers. The emails, the money and the rules behind the charge are public, and so are the gaps in the wider estimates.

Amazon Counts Over 1 Million Warehouse Robots, AutoStore About 87,500 and Locus More Than 22,000, While Symbotic Counts Systems and Ocado Counts Modules

Robots, systems, modules and sites are four different units, and each company picks the one that suits it. Every figure here comes with its source, its date and what it counts, plus the Indian warehouses that run on sorters.

Pakistan-Linked Hackers Spent Years Phishing India's Military. Now They Are Going After Its Universities.

A Windows shortcut with a PDF icon, a Linux file with a PDF name, a lookalike of ThePrint: the bait changes every few months. The targets, Indian government, military and now academic, have not.

Your AI Prompt Uses About Five Drops of Water. That Is Not the Number You Should Be Worried About.

Google says one Gemini prompt costs 0.24 watt-hours and five drops of water. Both numbers are real. Both are also close to useless for answering the question people are actually asking.

Indians Reported Losing ₹22,845 Crore to Cyber Fraud in 2024, and Nearly Half the Complaints Traced to Cambodia, Myanmar and Laos

A Lok Sabha reply put reported losses to cyber fraud at ₹22,845.73 crore in 2024. India has rescued 6,998 citizens from scam compounds abroad, ordered incidents reported within six hours and notified data protection rules, but its civilian cyber strategy is still unpublished.

The 2002 Roomba Cost $199.95 and Had No Map; Today's Humanoids Share a Charging Problem and a Lidar Supplier With the Robot Vacuum Industry but Little of Its Hardware

A 2002 Roomba navigated without a map because position sensing would have added over $1,000 to its cost. Documents from iRobot, Hesai, Roborock, Ecovacs and academic papers show which of its ideas reached today's humanoid robots.

1.45 Billion People Are Enrolled in One ID System. The World Bank Called It a Model for Every Country

A single twelve-digit number, attached to a fingerprint and an iris scan, now underpins how India banks the poor and pays subsidies straight into accounts. The World Bank's own chief economist called it the most sophisticated system he'd seen.

The World Built AI in English First. India Just Spent $1.25 Billion to Build One in Its Own Languages

A government program is renting out nearly 60,000 GPUs to Indian startups at a fraction of market price, and a translation platform fields tens of millions of requests a day in languages ChatGPT still stumbles over. Not every bet has paid off yet.

A Robot Hand Needed 13,000 Simulated Years to Learn the Rubik's Cube, and in 2026 a Five-Finger Hand Still Screws In a Light Bulb Only 36 Percent of the Time

From OpenAI's simulation-trained cube solver to DeepMind's 2026 five-finger hand, here is what each method has measured, which lab reported it, and where the success rates are still low.

India Runs the World's Busiest Payment System. No Company Owns It.

A vegetable seller takes nine rupees by QR code. It clears in two seconds and costs her nothing. Around 22 billion of those happen every month in India, on a system nobody is allowed to own.

The Four-Legged Robots Already Walking Into Places That Would Kill You

A quiet class of machine has slipped into some of the most dangerous workplaces on Earth: the radioactive room, the gas-filled mine, the collapsed building. They walk on four legs, and they don't need to breathe.

The Uncanny Valley Isn't a Law of Nature. It's a Design Decision

As humanoid robots pour into the world, an old dread is back: the queasy feeling when a machine looks almost human. We treat it as a fixed fact of the mind. It isn't. It's a trap engineers can choose to avoid.

2026 Is the Year Humanoid Robots Finally Clocked In

For decades the walking, talking robot was always 'a few years away.' This year they started showing up on factory floors and in airports, on real tasks with real hours logged. The catch is how few of them.

AI Just Won a Nobel Prize for Cracking a Problem That Stumped Biology for 50 Years

For 50 years, working out how a single protein folds took years of lab work. Then an AI learned to predict it in seconds, and it just won a Nobel Prize for it.

A China-Linked Group Targeted Four of India's Five Regional Grid Control Centres. Nobody Has Proved It Turned Off Mumbai's Lights.

Recorded Future found a China-linked group communicating with 10 Indian power organisations, four of them regional grid control centres. Whether it had anything to do with Mumbai's blackout is something Maharashtra, Delhi and the firm itself answered three different ways.

The Robot in Your Operating Room Doesn't Operate. A Human Does, for Now

Millions of surgeries a year are now done 'by robot.' But the da Vinci, the machine doing most of them, doesn't make a single decision. Every cut is a human hand, one step removed. The interesting question is what happens when that changes.

Your Chatbot Is Just Guessing the Next Word. So Why Is It So Good?

The model at the heart of it looks nothing up and knows no facts. It plays the most sophisticated game of autocomplete ever built, and that one idea explains what it gets brilliantly right and embarrassingly wrong.

Robots Ace Chess but Can't Fold Your Laundry. Meet Moravec's Paradox.

We assumed the hard stuff for machines would be logic and strategy. It's the opposite: the things a toddler does without thinking are what still defeat the world's best robots.

Amazon Just Passed One Million Robots. You'll Almost Never See One

Amazon crossed one million robots in its warehouses in June 2025, close in number to the humans working alongside them. Almost none of it is visible to a customer, which is exactly how the company wants it.

The Cabinet Approved 15,000 Spraying Drones for Women's Self-Help Groups in 2023, 500 Were Distributed by March 2026, and Field Trials Found Drone Spraying Matching or Beating Knapsack Sprayers

A Cabinet scheme for 15,000 drones, 500 distributed by March 2026, pilots trained at approved organisations, and three published field comparisons in which drone spraying matched or beat knapsack sprayers. The count in Parliament, the drone rules and one village report follow.

The New Bionic Limbs Don't Just Move When You Think. They Feel Back

For decades the dream of the bionic limb was control: a hand that moves when you think. But the deeper problem was the opposite: getting the limb to talk back to the brain. Researchers are finally cracking it.

Self-Driving Cars Were 'Five Years Away' for Fifteen Years. Then, Quietly, They Arrived

For fifteen years, fully autonomous cars were perpetually just over the horizon. Then, while everyone was being cynical, half a million people a week started climbing into cars with no driver.