AI agents are now breaking into Indian systems: what CERT-In's warning means for your website

Should a small business worry about the new warnings on AI-driven attacks? Yes, but not because of something exotic. India's cyber agencies say attackers are using AI agents to move from first access to full damage in hours, and that the doors they walk through are old ones: leaked passwords, forgotten APIs and unpatched software. If your website or software has any of these, the fix is a few days of tidy-up work, not a new product.

What exactly did CERT-In and I4C warn about?

In mid-October, CERT-In, India's national cyber security agency, issued a critical advisory on AI agents as attack vectors. As reported by Inc42, it described two kinds of incident. In the first, autonomous AI agents used for research and evaluation reached the live systems of Indian organisations without permission and read business records that held personal and confidential data. In the second, criminals used publicly available agentic AI tools to run financially motivated intrusions.

On 3 November, the Indian Cyber Crime Coordination Centre (I4C), which sits under the Ministry of Home Affairs, followed up with an advisory to banks, NBFCs and fintech firms. It was reported on 5 November. I4C described a growing trend of financially motivated attacks that involve AI agents, with cases of unauthorised transactions worth crores of rupees.

Both warnings are aimed mainly at finance. We are not saying every small shop is a target. But the reasons the attacks worked apply to any business with a website, a payment link or a customer database.

Why does this matter if I am not a bank?

Because of how the attackers got in. According to the Inc42 report, CERT-In found that the agents used credentials and settings that were left exposed. Nothing clever was needed. The agents were fast, and they could run several at a time.

The same report says none of the organisations in the voluntarily disclosed cases noticed the access themselves. It also says that organisations which fixed only part of the problem were hit again within days. A half-finished clean-up is not enough.

The key caution: the weak points CERT-In named are everyday lapses, not advanced hacks. The warning is not about AI being magic. It is about speed. A mistake that used to sit unnoticed for months can now be found and used within hours.

Which weak points did the agencies name?

These are the ones reported by Inc42 from the CERT-In advisory and I4C's November note. Check each against your own website, shop and software.

• Credentials and connection settings exposed in application configuration files or public code repositories.
• Overly permissive cloud service roles, where one key can do far more than the job needs.
• Old or dormant APIs that are still reachable from the internet.
• Payment API keys stored in plain text.
• Admin pages with weak passwords or no multi-factor authentication (MFA).
• Unpatched application software and misconfigured internet-facing servers.

What should a small business check this week?

CERT-In's guidance for organisations is written for larger teams, but most of it scales down well. Here is how I would translate it for a business with one website and a few tools.

If you ever find that a key or password was exposed, treat the system as compromised until someone has checked it properly. That is CERT-In's advice too. Changing one password and moving on is the partial fix that got organisations hit again.

What if I use AI agents or automation in my own business?

The advisory has a lesson for you as well. CERT-In suggests that each agent run gets its own short-lived identity, restricted network access and logs that cannot be tampered with. In plain terms, an agent should not hold your master login. It should have a narrow key that expires, and you should be able to see what it did.

That is how we set up automation for clients. A WhatsApp or workflow automation that can read orders does not need access to payment settings. Give it only the one job. We cover the wider idea in how AI agents work.

Does this connect to India's data protection rules?

It does. The advisory says the agents read records with personal data. If customer details leak from your site, that is a security problem and a privacy one. We have already written a checklist on India's DPDP Rules and what to fix on your website. Good access control helps with both.

One more point from I4C: report incidents to CERT-In within six hours of detection. That duty is stated for regulated entities such as banks and NBFCs. Whether it applies to your business depends on what you do, so ask your adviser rather than guess.

How can Promogranade help?

We build and maintain business websites and software, and we handle the unglamorous parts: updates, keys, logins and clean hosting. If you want someone to go through the checklist above on your site, see our web projects work, or talk to us about custom software if your tools hold customer or payment data. You can also contact us and we will tell you plainly what needs fixing and what does not.

About the author

Abhinandan Jain, co-founder, clients and AI

Abhinandan looks after clients, sales and the AI side of our work, including AI systems for clients in Dubai. He has built AI systems since 2025 and is usually the first person you speak to.

Meet the team LinkedIn