Every writeup of the DPDP Act 2023 leads with the same number: penalties of up to ₹250 crore. It's the right number to know, and it's also the most commonly misunderstood part of the Act. This is what it actually means, and what has to go wrong for it to apply.
We covered what the Act asks of AI tooling generally in an earlier post. This one is narrower: just the penalty mechanics.
It's a ceiling per violation, not a fine per breach
The Act's Schedule sets a maximum penalty for each category of violation a Data Fiduciary can commit. ₹250 crore is the highest tier in that Schedule — it isn't the number every violation carries, and it isn't a fixed fine the Data Protection Board of India hands out automatically. The Board determines the actual amount within that ceiling, based on factors the Act specifies: the nature and gravity of the breach, the type of personal data involved, whether the Data Fiduciary took steps to mitigate the harm, and their prior compliance history, among others.
So the practical question isn't "does this violation carry the ₹250 crore penalty" — it's "which tier of the Schedule does this fall into, and where in that tier's range does the Board land."
What specifically carries the top tier
The ₹250 crore ceiling attaches to a personal data breach caused by a Data Fiduciary's failure to take reasonable security safeguards to prevent it. That's a specific, narrower claim than "any data breach" — it's tied to the adequacy of the safeguards in place, not to the mere fact that an incident happened.
This distinction matters because it means the Board's inquiry isn't just "was there a breach" — it's "was this breach the kind that reasonable safeguards would have prevented, and were reasonable safeguards actually in place." An organisation that had a documented, functioning control and still suffered an incident is in a materially different position than one that had no control at all.
Other categories of violation — failing to notify the Board or affected individuals of a breach, failing to fulfil additional obligations around children's personal data, failing to give effect to a Data Principal's rights request — carry their own, lower ceilings in the Schedule.
"Reasonable" is judged after the fact, which is the uncomfortable part
The Act doesn't define a specific technical standard for "reasonable security safeguards" the way, say, a PCI-DSS requirement might specify particular controls. That's deliberate — a fixed checklist ages badly against changing technology — but it also means the Board evaluates reasonableness after an incident, against whatever safeguards were actually documented and operating at the time.
In practice, this rewards organisations that can produce evidence: a policy that existed before the incident, was actually enforced (not just written down), and generated a record of what happened. It's a much weaker position to argue "we had reasonable safeguards" without anything that demonstrates they were running, than to produce a hash-chained audit trail showing detection and enforcement decisions made in real time, before and during the incident window.
Where this connects to AI usage specifically
An employee pasting a customer database export into an AI chat tool is, in Schedule terms, a personal data breach if that data reaches a party outside any consent the customer gave. Whether the top-tier penalty applies then turns on whether "reasonable security safeguards" existed to prevent exactly that — and for most organisations today, the honest answer is that no control exists at the point a prompt is typed, because the tooling that would normally catch data leaving through files or email was never built to read a chat window.
That's the gap a browser-level control closes: not compliance in itself — nothing automates that — but the "reasonable safeguards" evidence a Data Fiduciary would otherwise have none of. See what PromptSec detects today, and what it doesn't yet, stated at the same size either way.
This isn't legal advice, and we're an engineering team, not a law firm — for how a specific incident would actually be assessed, talk to counsel.