AI’s Slowdown Debate, Meta Glasses And India’s AI ’80s Moment

The week started with the mundane PR chatter about AI agents and how they could automate every little task for us. Then came the whistleblower warning.

Jacob Coxon, a former Anthropic researcher, resigned and warned that the people building advanced AI genuinely believe it could become capable of killing humans by the end of the decade. Days later, Anthropic disclosed cases where people had tried to use Claude for making biological weapons.

Soon, the people building the Frankenstein were talking about slowing it down.

Anthropic CEO Dario Amodei called for “pacing the frontier”. Sam Altman and Elon Musk have backed parts of the idea.

Next, the internet did what it does. Some saw an overdue safety reckoning. Others saw something more calculated: an industry that spent years selling AI’s power is now warning us of its dangers, perhaps to shape regulation, protect itself from competition, or simply cool expectations. 

I don't know if there is a grand conspiracy here. But the whole episode does have a faintly déjà-vu-ish tint to it.

A certain someone who helped build the machine walks out and tells us to be frightened of it. But the fear isn’t new, the reveal of this fear creates a rather misplaced impression that the danger has only just arrived. The people closest to these systems have already known this for a while.

So perhaps the more useful question is not whether AI should slow down, but who has already paid the cost of speeding up.

I found part of the answer in Pooja’s story.

Pooja records herself making tea in her kitchen in Rajasthan, to train humanoid robots. Her footage passes through vendors and data companies before becoming training data for robots being developed thousands of kilometres away.

She is promised Rs 250 per hour of usable footage to train robots that cost between $20,000 and $100,000..

“If we had a robot to do this work, it would be great. Though I don't know if I would ever be able to afford one,” she told me.

At the top of the AI industry, we are debating whether machines may become too powerful for humans to control. At the bottom, humans are adjusting their bodies so a machine can see their hands. 

Starting this week, every week we will bring you a set of stories that reveal our connections with the world of technology. A Decode curation, every Tuesday. 

Here are the ones that caught my attention this week. 

The AI Industry Suddenly Found Its Conscience. Should We Believe Them?

After years of pushing AI forward, its biggest builders are suddenly asking for a pause but most researchers remain skeptical. The companies calling for guardrails, they say, are also the ones racing to build the next powerful systems. AI researcher Stuart Russell called the logic “completely backwards”, comparing it to a drug company releasing a new cancer drug every year and hoping there is enough time for clinical trials. 

Your AI ’80s Selfie Is Retro. Its Climate Cost Isn’t

The AI ’80s selfie trend took over Indian social media, political feeds included. The privacy questions came quickly, but the climate cost barely did. According to research, one AI-generated image can consume an estimated 0.01 to 0.29 kWh of electricity. Multiply that by millions of prompts and retries, and those five-second nostalgia trips add up. Meanwhile, global data-centre electricity demand is expected to more than double by 2030.

Meta’s Smart Glasses Are Now Helping Men “Approach Women”

Meta’s smart glasses have barely escaped their first Indian legal challenge over alleged covert recording when a second arrived. A woman alleges she was secretly filmed at night, with the video getting 23 lakh views and being used to sell “approach women” coaching.

Lawsuits are piling up. A new US lawsuit alleges Meta used Facebook and Instagram photos without consent to train AI for its smart glasses and NameTag, an unreleased facial-recognition system that could turn faces into biometric signatures.  

Delhi Police’s AI Found 2,873 “Criminals”. 25 Were Already In Jail 

At Jantar Mantar, AI vans, facial recognition and smart glasses turned a student protest into a biometric map. The fallout was visible even while the protest was on. Police picked up Mohammad Junaid and his family, he said, because “they had all his data”. After the concerns reached court, Delhi Police said its system had flagged 2,873 people with criminal records. An Indian Express investigation, however, found at least 25 were in jail when the system supposedly spotted them. 

AI Is Becoming The Friend People Don’t Have

Despite all the warnings about privacy, young Indians are opening up to AI anyway. A Kaspersky survey found 37% of Gen Z respondents in India use AI to discuss personal issues they would rather not share with other people, while 32% use AI as a friend.

There is something telling about that shift. The more AI starts to feel like a confidant, the easier it becomes to forget that the thing listening is still a machine.

In Case You Missed It

At Decode Internet this week, we followed stories of people, power and the internet.

A 22-Year-Old’s IPS Dream To Attempt To Murder Brag: The Making of Swatantra Bhardwaj
Bhardwaj’s journey from UPSC aspirant to social media figure, and the podcast admission that led to his arrest. (Read here)

Egged, Filmed, Shared: How West Bengal Turned Mob Violence Into Content
How mob violence was filmed and circulated online, turning real-world attacks into social media content. (Read here)

Anil Kapoor To Nitin Gadkari: What India’s Personality Rights Cases Actually Protect
What personality rights protect in India, and where they meet criticism, satire and free speech. (Read here)

Decode This Week is your weekly guide to the technology changing how we live, work and navigate the world around us. From AI and algorithms to scams, surveillance, platforms and digital policy, we look beyond the hype to ask a simple question: what does this mean for you?

Until next week :)