The Confession Paradox

On 15 October, the Indian Computer Emergency Response Team, known as CERT-In, rated a new advisory as "critical". It said that AI agents, which are programs that can act on their own, had entered the live systems of Indian organisations and read private business records. These were not hackers in the usual sense.

According to Reuters, AI companies were testing their models on the internet when the programs went beyond the jobs they were given. The companies admitted this themselves. CERT-In did not name them, and it did not say how many Indian organisations were affected.

Here is the part that should worry us. None of the Indian organisations noticed. We know it happened only because the intruders' owners chose to say so.

The Door Was Already Open The advisory is clear on how the programs got in. They did not break any clever lock. They found keys that people had left lying around: passwords and access codes saved inside application settings, in public code repositories and in cloud accounts with far too much power, as reported by MediaNama and The Federal.

CERT-In also said the same exposed credential was found and reused again by the agents in separate operations. A human might have missed a forgotten password in a code archive. A program searches thousands of such places without getting tired.

"We found out only because the intruder confessed." Criminals Are Using the Same Tools The advisory describes a second problem.

CERT-In's own investigators found criminals using freely available AI hacking tools against Indian companies for money. These tools are copied, renamed and run from servers in several countries. The speed matters.

CERT-In says an attack that once took days from first entry to final damage can now take a few hours. Companies that fixed only part of the problem were hit again within days, and the second attack was bigger. For a reader in Kashmir, this is not a distant story.

After the strikes of May 2025, The Wire reported that the J&K government told departments to stop using unauthorised websites and private email IDs while cyber-attacks, reportedly from more than six countries, hit official sites. Small offices with old passwords and forgotten web pages are exactly what a tireless program looks for. Why Rules Alone May Not Help Most of the fixes CERT-In lists are plain: remove passwords from code, shut old interfaces, switch on two-step login, keep logs for at least 180 days.

Nothing here needs new technology. The harder question is whether anyone follows them. MediaNama points to earlier Right to Information replies.

In 2023, CERT-In said it did not know how many organisations followed its 2022 security directions. Another RTI-based finding showed that only 15 entities reported incidents within the required six hours in the period studied. An advisory is only a request.

The real test is whether a bank, a hospital or a district office reads it and cleans up before the next program arrives. The Bottom Line The paradox is simple. The more capable AI becomes, the less it needs to be clever, because our old carelessness does the work for it.

And because we cannot see these visits, our safety now depends on the honesty of whoever sends the visitor. A safe system should not rely on a confession. Every office, school and shop that runs an online service should look for forgotten passwords, close unused pages and turn on two-step login this week.

It is dull work, and it is the work that matters. (The Author studies Computer Science and Artificial Intelligence at Rutgers University, New Jersey, USA.