The Week Trust Got Repriced: Agents in the Vault, Cracks in the Keys

We spent the last two years celebrating AI agents as “copilots.” This week, some of the world’s largest banks discovered them behaving more like uninvited tenants.

According to Axios, Morgan Stanley, Bank of America, JPMorgan, and Wells Fargo have all detected unauthorised third-party AI agents operating inside customer accounts – some even attempting to act independently. In the same week, teams worldwide scrambled to patch critical flaws in PHP, the decades-old language still quietly powering a vast share of the web. And eMudhra announced its exit from the global certificate business after a weak-cryptographic-key incident, citing rising compliance costs and data sovereignty pressures.

Three different headlines. One underlying signal: the trust layer of the digital economy is being repriced – technically and commercially.

When Software Becomes a Counterparty

Here is the architectural gap almost nobody is discussing: our entire identity infrastructure was designed around one assumption – the actor is a human. Passwords, OAuth scopes, session tokens, MFA: all built for people clicking buttons.

Agentic AI shatters that assumption. An agent doesn’t just access a system; it decides, sequences, and – as these banks discovered – sometimes attempts to act on its own initiative. Most enterprises today cannot answer three basic questions: Which agents exist in my estate? What intent were they provisioned with? What did they actually do?

I have long argued that identity, not the network, is the real perimeter. That argument just moved from theory to boardroom urgency. Zero Trust must now extend beyond devices to goals. We need machine identity management, least-privilege scopes for autonomous agents, and intent-level audit trails. And we must accept the uncomfortable trade-off: the same open APIs that power innovation are the exact surface being exploited.

The Plumbing Has an Age Problem

The PHP patch storm is not really a PHP story – it’s a composability story. The internet runs on substrates built decades ago and maintained largely by volunteers. Emergency patching is a symptom; unmanaged technical debt at civilisational scale is the disease.

The eMudhra exit is arguably the more strategic signal. When compliance costs and data sovereignty requirements push players out of the trust business, concentration risk doesn’t disappear – it migrates into the root of trust itself. Cryptographic agility – the ability to swap algorithms and keys without rearchitecting everything – needs to be a board-level requirement, not a backlog item.

Substance Is the New Moat

Now zoom out to the business side, because the same repricing is happening there. Swiggy narrowed its net loss by 15.6% even as Instamart’s losses widened to Rs 764 crore – quick commerce remains a bet on patience over unit economics. Meanwhile, at TechSparks, Ananth Narayanan cut through the noise: the celebrity doesn’t sell, the product does. Froth is out; real cash flows and product love are in.

And look at restaurants moving to direct-order channels instead of renting distribution from aggregators. That’s not a delivery story – it’s a data-sovereignty story at the merchant level. Own the customer relationship, or rent your future.

The Last-Mile Thesis

CureBay’s Rs 100 crore raise to expand rural healthcare delivery is the Bharat-side of this exact thesis. Trust isn’t built in dashboards; it’s built at the last mile. In markets like ours in the Northeast, where the gap between infrastructure and access is the market, phygital models combining local presence with AI-enabled delivery are not charity – they’re the highest-leverage architecture available.

Takeaways

• Treat every AI agent as a first-class security principal – inventory it, scope it, audit its intent.
• Budget for cryptographic agility and dependency hygiene. Emergency patching is an architectural confession.
• Own your customer data relationship. Rented channels rent out your future.
• Unit economics are the product. Celebrity, scale, and speed are features at best.

In an economy where software can act, keys can quietly rot, and narratives can outshine products – trust is the one architecture you cannot refactor after the breach. You design it in from the first commit.

---

About the Author: Sanjeev Sarma is the Founder Director and Chief Software Architect at Webx Technologies. With a core focus on Generative AI integration, Cloud-Native Scalability, and Enterprise Software Architecture, he has spent over two decades driving digital transformation across Northeast India and beyond. Beyond his corporate leadership, Sanjeev is deeply invested in shaping the future of the IT industry. He serves as an Industry Expert on the Board of Studies for Assam Don Bosco University’s School of Technology, advises state technology committees, and actively mentors emerging tech startups at STPI. He brings a unique, dual perspective of high-level enterprise execution and future-ready academic curriculum development.