Summary

  • The CERT-In directions dated 28 April 2022 give an organisation six hours from the moment an incident is noticed or brought to notice — a clock that measures what a company knows, not what an attacker did.
  • The same instrument reaches service providers, intermediaries, data centres, bodies corporate and government organisations and demands 180 days of logs kept inside Indian jurisdiction, yet across the two official-domain documents retrieved for this briefing it publishes no register, no count and no enforcement record that an outsider could use.

Six hours is one meeting and two phone calls. It is also, on the reading of the CERT-In directions retrieved for this briefing, the whole window India gives a service provider, an intermediary, a data centre, a body corporate or a government organisation to tell the national computer emergency response team that it has been compromised. No other general instrument I have read in this beat sets a shorter one.

The directions were issued under section 70B(6) of the Information Technology Act, 2000 and are dated 28 April 2022, and the ministry's own copy of them sits on a government file path, which is where the CERT-In cyber security directions of 28 April 2022 were retrieved on 16 November 2026. Everything this briefing draws from that document is a summarised reading of the copy retrieved that day; I did not get a verbatim sentence out of it, and the article says so wherever the difference matters.

The first thing worth understanding is where the clock starts. The period runs from the incident being noticed, or being brought to notice — not from the moment the intrusion began. That distinction is the difference between a deadline a lawyer can compute and a deadline a duty-holder sets for itself. An attacker who has been inside a network for five months has caused exactly nothing as far as this clock is concerned until somebody internally sees it or an outsider tells them. From that instant the organisation has six hours, which is enough time to convene a bridge, freeze an account and decide who calls whom.

It is not enough time to complete an investigation, and the instrument does not pretend otherwise: the first report is a provisional statement of what is known, not a finding.

Read as a governance choice rather than a formality, that trigger is unusually honest. Most incident-reporting regimes shelter behind an objective event, which lets the reporter argue about when the event really occurred. CERT-In's wording puts the argument inside the organisation, where it belongs, and makes delay a question about when the organisation allowed itself to know.

The second thing is who the direction is aimed at. The summarised reading of the retrieved document lists service providers, intermediaries, data centres, bodies corporate and government organisations. That is a class wide enough to catch nearly every organisation in India that touches a network, from a two-person hosting reseller to a ministry. It is worth pausing on what an auditor can and cannot do with a list like that.

On paper it means no one gets to say the rules were not meant for them. In practice it means the enforcement question is severe, because a rule that reaches everyone is a rule a regulator cannot police one case at a time. The directions do not appear, in the material I retrieved, to solve that problem: there is no register of reports, no aggregate count and no enforcement record in either of the two documents I read, so the only people who can currently tell a six-hour report from a three-week-late one are the two parties to it.

That gap is not a technicality. A deadline gains its force from the possibility of being checked, and the check here would be trivial to build: a quarterly figure, a table of reports received against reports due, one adjudicated outcome. Instead the instrument is exacting at the reporting end and silent at the public end. An organisation that has been breached carries the same outward appearance whether it filed at hour five or hour five hundred. The Indian Computer Emergency Response Team may well hold that record internally. Nothing I retrieved shows it, and a consistent absence across a small set of documents is not the same thing as a finding about what an agency knows.

The third element is the part boards tend to under-read, because it is not about notification at all. The summarised reading of the same document requires logs to be kept for a rolling period of 180 days, inside Indian jurisdiction. Six months of retained history is a statement about infrastructure: you cannot comply by writing a good report, and you cannot buy compliance in the week after an incident. Either the retention was already there, with the storage, access control and integrity work that makes logs usable as evidence, or it was not.

The jurisdiction requirement compounds it, because an organisation whose logging pipeline terminates in another country has to decide whether to move the pipe or duplicate it.

The practical consequence is that the six-hour clause and the 180-day clause punish different failures. A late report is a process failure that can be argued about after the fact. A missing log is an architecture failure, discovered at exactly the moment it can no longer be fixed, and it converts a recoverable incident into an unreconstructable one. Of the two, the retention duty is the one that quietly governs whether the six-hour report is worth anything: a report filed at hour five from an organisation that cannot reconstruct the preceding week is a notification, not intelligence.

The reporting channel described in the retrieved copy is unglamorous and appropriate — electronic mail, telephone, or a web portal. Cheap to use, easy to audit if anyone chose to. The annexure of reportable incident categories is the part I hold least well, and it matters, because a duty is only as tight as its trigger list: a broad definition pulls in noise, a narrow one invites the argument that this particular compromise was not on the list.

The statutory frame around all of this is section 70B of the Information Technology Act, 2000, available on India Code's page for section 70B and retrieved the same day. The summarised reading of that page says section 70B provides CERT-In's statutory basis and that subsection (7) addresses non-compliance with directions issued under subsection (6). It also renders subsection (7) as imprisonment up to one year, or a fine up to one lakh rupees, or both.

I am deliberately not asserting that as the current price of missing the clock. That wording is what a summary says the page says, I never read a verbatim provision, and India has spent recent years converting exactly this kind of criminal sanction into an adjudicated civil penalty across a wide range of statutes. A summary can lag; a statute can change; a figure that was right in 2000 is not evidence of what a compliance officer faces now.

So the honest form of the claim is narrower and still useful: the directions sit on a statutory provision that contemplates a consequence for non-compliance, and the material I hold does not establish what that consequence currently is.

What this adds beyond the coverage that already exists is a different question about the same institution. A briefing published on 16 October 2026 took the AI-evaluation angle — model providers reporting that testing agents had reached Indian systems, and affected organisations not detecting the access themselves. That is a piece about discovery.

This one is about the instrument that would have to be complied with once discovery happens, and about the fact that the instrument's strongest feature and its weakest are the same design decision: it asks the duty-holder to grade its own homework and then files the result where nobody else can read it.

A board reading this should take away three things, in order. The clock is real and it starts when you know. The log retention is the harder obligation, because it cannot be met late. And the accountability runs one way: outward to the state within six hours, inward to nobody at all. If a register, a count or a single adjudicated outcome were ever published, that last sentence would change — which is precisely how a reader can check whether anything has moved.

Three limits travel with everything above. Across the two source snapshots retrieved on 16 November 2026 and the seven research requests made in this run between 19:55 and 20:00 UTC, no publisher-verbatim passage from either document was obtained; both retained snapshots are summaries whose excerpt field is empty. The exact trigger sentence, the full annexure of reportable categories and any consequence stated within the directions themselves are therefore not held in the documents' own words.

And whether the directions of 28 April 2022 remain in force unamended as of 16 November 2026 was not established, because no supersession instrument was read or located in the material reviewed.