When AI starts spending your money

Our goal with The Daily Brief is to simplify the biggest stories in the Indian markets and help you understand what they mean. We won’t just tell you what happened; we’ll tell you why and how too. We do this show in both formats: video and audio. This piece curates the stories that we talk about.

You can listen to the podcast on Spotify, Apple Podcasts, or wherever you get your podcasts and watch the videos on YouTube. You can also watch The Daily Brief in Hindi.

In today’s edition of The Daily Brief:

1. When AI starts spending your money

Why are NPCI and card networks building frameworks for agentic payments? While AI can already compare products and find deals, completing purchases currently requires human handover for approval. To automate the final checkout, systems like UPI Circle and Reserve Pay are creating delegation protocols that set strict spending limits and digital identity checks. This allows AI agents to buy products on your behalf while protecting account details and managing liability risks.

2. How pure is your silver, really?

Why is the Indian government moving toward mandatory silver hallmarking? Unlike gold, silver purity has lacked compulsory regulation, leaving buyers reliant on unverified jeweller stamps. BIS is expanding its Hallmark Unique Identification (HUID) system to silver, stamping articles with traceable digital IDs. While testing infrastructure and transport costs pose hurdles for small jewellers, the mandate aims to bring standardised purity and consumer protection to India’s fragmented silver market.

Ask an AI to find you a good pair of shoes today, and it can already do most of the annoying work. It can search across websites, compare prices and narrow down the options. But eventually, you still have to take over. You open the merchant’s app or website, add the shoes to your cart and approve the payment.

Companies are already trying to remove that final handover. Razorpay and the National Payments Corporation of India, or NPCI, have piloted purchases inside ChatGPT and Claude using UPI tools such as UPI Circle and Reserve Pay. Pine Labs has also built its own protocol on top of UPI. Mastercard and Visa are building similar systems for card payments, while Google has proposed a broader standard called AP2.

These are separate products, pilots and protocols built by individual companies or partnerships. India may now try to create a common framework across the UPI ecosystem.

Reuters reported that NPCI is working on a Unified Agent Protocol, or UAP. It could give banks, merchants, payment apps and AI agents a shared way to identify an agent, define what it may spend, record what it did and decide who is responsible when something goes wrong.

Until now, software mostly helped move money after you had chosen what to buy. Agentic payments ask it to do both: make the choice and complete the payment.

Before we get to AI, consider what happens during an ordinary payment. The system must know where the money should go and whether you approved sending it.

With cash, both are obvious. If you buy something for ₹500 and hand the seller a ₹500 note, the money changes hands because you physically gave it to them. No bank or payment network needs to separately verify your permission.

Digital payments work differently. The instruction to move money and the proof that you approved it are two separate things.

One instruction tells the system who should receive the money and how much. Another check establishes that the account holder approved it.

A cheque does this using your signature. In a card payment, the merchant sends a request through the payment network to the bank that issued your card. The bank checks the transaction and either approves or rejects it.

Internet banking lets you send instructions directly from your bank account. IMPS made these transfers instant and available around the clock. These systems did not neatly replace one another. Each removed a different piece of friction.

Then UPI compressed much of the process into a few taps.

UPI launched in 2016 and gave banks and payment apps a common system. Instead of entering someone’s account number and IFSC code, you could scan a QR code or enter a UPI ID.

The QR code or UPI ID tells the app who should receive the money. You enter the amount, choose your bank account and approve the transaction. UPI then carries that instruction between the apps and banks involved.

But the bank still needs to establish that you authorised the payment. That is what the UPI PIN helps do. The PIN does not unlock money stored inside your phone. It tells your bank that you approved that particular instruction.

From the user’s point of view, a digital payment therefore involves two basic questions: where should the money go, and did the person controlling the account agree to send it?

UPI made answering those questions remarkably easy. In August 2026 alone, Indians used it for 24.51 billion transactions worth ₹29.82 lakh crore.

But the person generally still had to approve each purchase. That began to change with mandates.

UPI AutoPay allows you to approve payments to a known merchant in advance. There is nothing particularly intelligent about telling software to pay Netflix ₹499 every month. The merchant is known, the purpose is known and the rule is largely fixed.

The software is just executing your decision, and not really making one for you.

Now let’s change the instruction. Tell the software to keep you subscribed to whichever streaming service offers Formula 1 at the lowest price.

The payment is suddenly the easy part. The software must figure out which service has the rights, compare the plans and decide which one offers the best deal. It may even need to cancel one subscription before starting another.

Large language models, or LLMs, have made it easier for software to work with the vague way people naturally speak. You can ask for a “decent morning flight to Delhi that isn’t too expensive” without defining exactly what decent, morning or too expensive means.

That does not mean the model will interpret those words exactly as you intended. It simply means the software can now attempt to work with them.

Give the model access to other tools and it can do more. A search tool can pull live flights. A calendar can tell the agent when you need to arrive. A browser can fill in the booking details. Add a payment tool, and the agent can carry the task through to checkout.

But it cannot simply take your card or bank account and spend whenever it wants. The payment system needs a way to limit that authority.

Card-based systems are trying to solve this using tokens and records of user consent.

Suppose you ask an agent to book a flight to Delhi for no more than ₹10,000. You may first authorise it to use a particular card within defined limits.

The payment provider does not need to give the agent your actual card number. It can create a separate digital credential linked to that card. This substitute credential is called a token.

The token may be restricted to a particular merchant, amount or period. The agent can use it after finding a flight that fits your instructions, while the payment system checks whether the transaction falls within the authority you granted.

The token protects the card details. Other parts of the system still need to record what you asked the agent to do. After all, proving that the agent had a valid payment credential is not the same as proving that it followed your instructions correctly.

UPI approaches the problem differently.

One building block is UPI Reserve Pay. It allows a customer to approve a block of money and lets multiple payments be taken from that block without requiring a new PIN each time.

Suppose you block ₹5,000 for purchases from a grocery platform. That money is set aside in your account. The platform can then collect several smaller payments from it as orders are placed, subject to the permission you gave.

This gives an agent room to make repeated purchases, but only from a limited pool of money and generally for a specified merchant. Reuters reported that banks currently cap such blocks at ₹10,000 for up to 90 days, although those limits may be reconsidered for agentic payments.

The second building block is UPI Circle.

UPI Circle was originally designed to let an account holder delegate limited payment authority to another person. Under full delegation, the second person can complete payments from the primary user’s account without asking for approval each time, as long as the transaction stays within the limits set by the account holder.

NPCI has since extended this idea to certain connected devices and software. An approved device or software profile can act as the secondary user and make domestic merchant payments within defined limits.

This is not open access for any chatbot that asks nicely. Under the current rules, the software or device must be approved, linked with the user’s consent and checked during each payment request. Such payments are currently restricted to domestic person-to-merchant transactions, with caps on individual transactions and monthly spending.

Reserve Pay and UPI Circle therefore solve different parts of the problem. Reserve Pay limits the pool of money available. UPI Circle decides who—or what—has the authority to initiate the payment.

The reported Unified Agent Protocol could tie these pieces together. It may give merchants a common way to connect with AI agents, allow users to express rules about when and how much an agent may spend, and create identity checks and audit trails for agent-led transactions.

Instead of every AI company, merchant and payment provider building its own arrangement, they could work through a shared set of rules. But getting the payment side ready solves only one part of the transaction. The agent must also understand what merchants are selling.

Think about how you shop online. You can usually make sense of a messy product page because you know how to fill in the gaps. You can spot that two differently named products are essentially the same, judge whether an old review is still useful and work out whether something will reach you on time.

An agent has a harder time doing that.

It works best when important details such as price, size, stock, delivery time, return policy and product specifications are presented in a structured form.

A 2026 study by Visa Acceptance Solutions reported that 48% of online shoppers had used AI while researching their most recent purchase. But only 15% of merchants had catalogues that agents could easily discover and interpret.

Consumers may start shopping with AI before many merchants are ready to serve AI shoppers.

That could quietly favour large marketplaces. Amazon and Flipkart already organise prices, reviews, stock levels, delivery estimates and return policies in one place. Smaller merchants may have the same product at a better price but present the information in a form that an agent finds harder to read.

So even if an agent can search across the internet, it may lean towards marketplaces simply because they are easier to compare. That is a possibility, not yet an established outcome.

Suppose you ask an agent to book a “reasonably timed” morning flight to Delhi for under ₹10,000. It finds one for ₹9,550 and books it. The fare fits the limit and the payment goes through. But the flight leaves at 4:45 am, while you meant something closer to eight.

Nothing was stolen. The agent had permission to spend and followed the stated budget. It simply interpreted “reasonably timed” differently.

Agentic payments can therefore go wrong in at least three ways. An agent may spend without valid permission, making it an unauthorised transaction. It may have permission but exceed the limits attached to it, suggesting that someone failed to enforce them. Or it may stay within every rule and still make a poor choice. In that case, the payment worked; the judgement failed.

That judgement may also be shaped by incentives the user cannot see. The company operating the agent decides which merchants it can access and may have commercial relationships with some of them. An assistant could recommend a ₹9,000 product from a marketplace that pays it, even when an equally good ₹8,500 option exists elsewhere. The purchase is valid, but the agent may be serving two masters. Economists call this the principal-agent problem.

Someone could also manipulate the agent directly. Agents read websites, emails and product pages while they work. A harmful page or message could feed them misleading instructions and influence what they choose. In that case, a fraudster may not need to steal your PIN at all. They only need to manipulate the software that already has permission to act for you.

Google says this kind of prompt injection is one of the biggest risks for agentic browsers because it can push an agent into actions the user never intended, including making a payment.

So who absorbs the loss? Existing rules may cover payments made without authority. If an agent exceeds a clear limit, responsibility may lie with whoever failed to enforce it. But if it follows every rule and simply buys the wrong thing, reversing the payment may not be enough.

NPCI’s reported framework is expected to address liability, but no details are public. Until they are, people may delegate grocery orders—but think twice before handing over flights, investments or expensive purchases.

If you’ve ever bought a silver anklet or chain, you probably looked at the “925“ stamp and took the jeweler’s word for it. It is ideally supposed to mean that in the article, at least 925 parts out of every 1,000 are silver.

How do you know this for sure? Usually, such a piece gets a hallmark from an independent party, which indicates that this number is verified.

But there’s one problem. Unlike gold, where every new piece sold in a covered district must carry a digitally traceable BIS hallmark, silver has had no such mandatory requirement. So a jeweller could stamp any number they liked, and it was the customer’s prerogative to believe them at face value or not.

The government plans to begin mandatory silver hallmarking soon. While BIS’s own mandatory-hallmarking page doesn’t list silver yet, BIS has already been putting in guardrails. Since September 2025, every silver article that a jeweller voluntarily gets hallmarked receives its own six-character digital ID which is traceable. More than 77 lakh silver pieces have gone through it so far.

So, how does hallmarking work?

The jeweller sends the article to a BIS-recognised Assaying and Hallmarking Centre (or AHC), an independent lab equipped to test precious metal purity. If the metal matches the declared fineness, it gets a BIS hallmark. For now, BIS hallmarking in India covers only two precious metals, gold and silver.

But, at the end of the day, a hallmark is a stamp that can be forged. Anyone could use it to deceive a customer.

That’s why BIS introduced the six-character digital code we mentioned above to all hallmarked articles. It’s called a Hallmark Unique Identification, or HUID.

Even before HUID, BIS hallmarks already identified the purity, testing centre and jeweller. But that information was only physically stamped on the article. There was no centralised digital record a buyer could independently check.

HUID adds unique article-level data that buyers can look up in the BIS CARE app. Enter the code and you can see the purity, article type, hallmarking date, testing centre details and jeweller registration number. That makes the certification far easier to trace, though it doesn’t perform a fresh test of the metal by itself. BIS has already tested this model with gold.

All of this is fine. But we couldn’t help asking ourselves: why was compulsory hallmarking for silver only being considered now?

Well, it turns out that gold itself remained voluntary for two decades. That’s why, by 2021, only about 30% of gold jewellery in India was hallmarked. That same year, the mandatory order was implemented, but initially, it was only in districts where testing centres were available.

Gold HUID was introduced just a month later. From April 2023, the government restricted the sale of older-format hallmarked gold without HUID.

Silver has also moved through similar stages on similar timelines. Silver’s hallmarking problem is also bigger than gold because silver articles are cheaper, sold in far higher volumes, and spread across a much more fragmented market. That makes the case for compulsion clear.

But passing the order is the only straightforward thing about it. The implementation is much harder.

You see, for businesses submitting new stock, the jeweller has to submit a batch through BIS’s system, send the articles to an AHC, wait for them to be tested and marked, and then collect them. Retailers buying already-hallmarked stock from manufacturers don’t necessarily manage this process themselves, but they have to pay those who do.

For a large manufacturer in a major city, this is manageable. But for a small jeweller selling anklets, toe rings and diyas in a town far from the nearest AHC, it is a tedious and expensive matter.

The prescribed hallmarking charges are ₹35 per silver article, subject to a minimum of ₹150 per consignment, plus applicable taxes. Gold costs ₹45 per article, with a ₹200 minimum. Jeweller registration with BIS is free. ₹35 is barely noticeable on a ₹10,000 article, but it’s a different proposition on a ₹500 toe ring. The existence of a minimum charge also disadvantages small consignments.

Beyond that, someone has to transport valuable metal safely, and lower-value silver articles come in huge volumes. That security service also needs to be paid for. All of this is most burdensome on small, less-capitalized jewellers.

Additionally, there was a reason the government only started gold hallmarking with districts that already had testing facilities. Silver is likely to face the same infrastructure constraint.

In September 2025, India had roughly 230 silver AHCs across 87 districts. By August 2026, that had grown to 341 centres across 102 districts, with more than 25,000 silver jewellers registered. That is fast growth, but India has nearly 800 districts. Gold’s mandatory system, by comparison, has enrolled over two lakh jewellers. Some gold AHCs can theoretically test silver, but to qualify for both, an AHC has to apply for further BIS certification.

More than that, silver appears in places besides jewelry, like utensils, religious articles, gifts and so on. India’s annual silver consumption is estimated at 5,000-7,000 tonnes, with jewellery accounting for about 35%. India’s gold jewellery demand, for comparison, was 430.5 tonnes in 2025. A mandatory regime doesn’t just mean testing more metal, but also processing millions of low-value individual objects spread across a highly fragmented supply chain.

That being said, numbers suggest jewellers are already moving into the system. Around 32 lakh silver articles were hallmarked in FY25. That rose to roughly 59 lakh in FY26. Since HUID was introduced, more than 77 lakh articles have received one.

None of this becomes real until the government publishes the final order for making silver hallmarks compulsory.

Again, gold is a useful guide here. It gave exemptions to small jewellers, very lightweight articles, and certain traditional jewellery. If you already owned un-hallmarked gold when the mandate came in, you could keep it, wear it, and resell it to a jeweller. The mandate applied to what jewellers sell to customers, not to what consumers sell back to jewellers. Nobody had to rush to get their old jewellery tested.

For silver, similar questions stand. Will the mandate cover only the 102 districts that currently have silver AHCs, or go wider? Will tiny articles, utensils and puja items be treated differently from jewellery? Will consumer-owned household silver get the same treatment? Any silver notification will need to be checked for how it handles these.

– This edition of the newsletter was written by Manie & Vignesh

1. Auto dealers face service revenue pressure as alternative-fuel vehicles overtake petrol cars

Petrol passenger vehicles fell below the combined share of EVs, hybrids, and CNG vehicles in August for the first time. Dealers are now discussing ways for manufacturers to offset lower after-sales income as EVs require less routine servicing.

2. SEBI eases compliance rules for FPIs investing only in government securities

SEBI has removed the requirement for foreign portfolio investors that invest exclusively in government securities to furnish investor-group details. The change follows the RBI’s withdrawal of concentration limits for such investors, making the earlier disclosure requirement redundant.

Source: The Economic Times

3. REC raises ₹500 crore through India’s first tokenised corporate bond

State-owned REC has issued India’s first tokenised corporate bond, raising ₹500 crore through a blockchain-based platform operating under the Securities and Exchange Board of India’s regulatory sandbox. The issuance used a tokenised settlement framework and same-day execution mechanisms to enhance efficiency and transparency in capital markets.

Source: The Hindu BusinessLine

4. India’s crude import reliance remains concentrated despite adding new suppliers

Despite adding 10 new crude oil suppliers recently, India’s import diversification remains limited, with the top five countries still accounting for 76% of total inbound shipments. Nine of the new suppliers contributed just 1.4% of total imports, while returning supplier Venezuela accounted for nearly 5%.

Source: The Economic Times

5. SECI prepares pooled green-power framework for MSMEs and industrial heating

The Solar Energy Corporation of India is working on a framework to aggregate green-power demand from MSMEs and industrial heating users. By pooling demand from smaller buyers, the scheme aims to make renewable electricity more affordable and easier to procure for industries looking to reduce their use of fossil fuels.

Source: Financial Express

We sat down with Mudit, co-founder of Voltseal, to talk about the duck curve, the many problems batteries can solve, and what happens when the grid finally gets intelligent. With over two decades in the industry, he helped us connect what we had learned through research with how the power system actually works.

Watch the full episode on YouTube.

There’s always more happening at Markets by Zerodha.

• The Chatter: How is SEBI targeting network-level market risks, and why is Maruti Suzuki betting on biogas alongside EVs? Plus, key updates from United Breweries, Indian Bank, Zen Technologies, and Lumino.
• Aftermarket Report: Why did Brent crude near $100 drag Nifty down to 23,635? Strikes on Saudi facilities raised rate hike fears. Meanwhile, Reliance announced a ₹12,500 Cr bond issue, and the UK exempted Indian carbon exports from double taxation. Plus, updates on HCLTech, PVR INOX, and Huawei.
• Points & Figures: How did India’s mutual fund industry surge to ₹74 lakh crore as SIPs became a monthly habit for millions? What is driving the rise of passive investing, and how far has the boom spread beyond top cities?

Join us on WhatsApp, where we share interesting soundbites from concalls, articles, and everything else we come across throughout the day. You’ll also get notified the moment a new video or article drops so that you can read or watch it right away.

This post was first published on Substack.

Thank you for reading. Do share this with your friends and make them as smart as you are 😉