Data Security

A data breach, hour by hour: the first 72 hours from the inside

A data breach, hour by hour: the first 72 hours from the inside

7 min read·1635 words·Updated 22 September 2026

The call comes at 6:40 on a Sunday morning, because it always does. A monitoring alert has fired, or a customer has emailed a screenshot of their own data on a forum, or, in the case I keep coming back to, a junior analyst noticed that a service account was logging in from a country the company doesn’t operate in. What happens in the next 72 hours decides most of the outcome: the cost, the regulatory result, the reputational damage and whether the people involved are still employed in a year. I’ve lived through this from inside a bank’s operations centre and, later, as the consultant they ring when they don’t have one. This is the timeline as it actually runs, not as the playbook says it should.

Hour zero to two: confirm, don’t react #

The first instinct is to pull the plug. Shut the server, reset the passwords, block the IP. Sometimes that’s right. More often it destroys the evidence you’ll need to work out what was taken and tips off an attacker who’s still inside that they’ve been seen, at which point they either accelerate or go quiet and come back through a door you haven’t found.

The first two hours are for confirmation. Is this real? What system? What’s the scope of access that account had? Get one person who can read logs and one person who can make decisions into the same call and keep everyone else off it. Start a log, a plain text document with timestamps, of every observation and every action, because in three weeks a lawyer, a regulator and an insurer will each want to know exactly what you knew and when. Memory is not evidence. The log is.

Free password lock laptop image — data breach response
Photo: Rawpixel via Rawpixel (CC0 1.0)

Hour two to twelve: contain, preserve, escalate #

Once you know it’s real, containment is the job, and it’s a set of narrow actions rather than a big one. Disable the compromised credentials. Isolate the affected hosts from the network without powering them off, so the memory and disk are intact for forensics. Snapshot everything. Rotate the secrets that account could have reached, and assume it reached all of them. Check whether the attacker’s foothold extended to the backups, because if it did your recovery plan is now their leverage.

Escalation runs in parallel. The general counsel, the chief executive, the cyber insurer’s hotline (read the policy; most require notification within a short window and will send their own responders), and the Australian Cyber Security Centre through cyber.gov.au, which is both a reporting obligation for some entities and, in my experience, a genuinely useful source of intelligence about who you’re dealing with. If the incident involves ransomware and a payment is being considered by anyone, stop and read the next section before that conversation goes further.

The mistake made most often in this window is the executive who wants to send a reassuring email to customers before anyone knows what happened. The reassuring email is wrong in a measurable fraction of cases, and a company that said “no customer data was affected” on Monday and corrected it on Thursday has a worse problem than the breach.

Hour twelve to 48: the assessment clock is running #

This is where the law arrives. Under the Notifiable Data Breaches scheme in Part IIIC of the Privacy Act, an organisation that suspects an eligible data breach has 30 days to assess whether it is one, and must notify the Office of the Australian Information Commissioner and the affected individuals as soon as practicable once it concludes that it is. Thirty days is the maximum, not the target. The OAIC’s guidance is explicit that the assessment should be expeditious, and its published enforcement actions show a clear pattern: the organisations that get the harshest treatment are the ones that sat on a known breach.

The assessment question is whether the breach is likely to result in serious harm to any of the individuals whose information was involved. That turns on what data, how much, who has it, and whether it was encrypted in a way that actually holds. It doesn’t turn on whether you’d prefer not to notify. Get legal involved in the assessment from the start and document the reasoning, because “we concluded it wasn’t serious” is a defensible position only if the conclusion was reasonable at the time and written down.

Separately, if the organisation has a turnover above the threshold set in the Cyber Security Act 2024 and makes a ransomware payment, or someone makes one on its behalf, that payment must be reported to the Department of Home Affairs within 72 hours. That rule commenced in 2025 and it changed the conversation in the room: the payment is no longer a private decision, and the people advising on it know that.

A breach, hour by hour. The first 72 hours and the clocks that start running
Diagram: Technology Spectator

Hour 48 to 72: scope, and the second breach #

By the third day the forensic picture is usually clear enough to say what the attacker touched, and it’s usually worse than the day-one estimate and better than the day-one panic. This is when the organisation finds out whether its logs went back far enough. A 30-day log retention on the identity provider means an intrusion that started on day 31 has no beginning, and the assessment has to assume the worst. I’ve had that conversation with a board. It’s short and nobody enjoys it.

The second breach is the one you cause yourself in the recovery. Restoring from a backup that was taken after the attacker got in. Rebuilding a server with the same misconfiguration that let them in. Announcing a “full reset” of customer passwords with a link that trains every customer to click links in emails about the breach, which the attacker’s phishing crew will imitate within hours. Recovery needs the same discipline as containment, and it’s the phase where tired people make the expensive errors.

The notification itself #

When the assessment concludes that individuals must be told, the notice has a required shape: what happened, what information was involved, what the organisation has done about it, and what the individual should do. The OAIC publishes a form for its own notification and guidance on the individual one. Write it in plain language, at a reading level a tired person can follow on a phone, and resist the legal team’s instinct to make it say nothing. A notice that reads “we take your privacy seriously” and then declines to say what was taken generates more complaints, more media and more regulatory attention than one that says “your name, email and date of birth were accessed; your password was not; here’s what to watch for”.

Two practical points. Send it from a domain the customer recognises, not a third-party mailing platform’s, because the attacker is about to send a fake version and you want yours to look more legitimate than theirs. And staff the phones. A notification that goes to 200,000 people at 5pm on a Friday with a call centre that closes at 5:30 is a self-inflicted wound I’ve watched happen twice.

The week after #

The incident doesn’t end at hour 72, but the character of it changes. Forensics finish. The regulator’s questions arrive, and they’re detailed: the timeline, the controls in place, the controls that failed, the assessment reasoning. The insurer’s questions arrive, and they’re about whether the policy’s conditions were met. The board wants a root cause, and the honest root cause is usually a combination of one missing control and one process that everyone knew was weak.

The organisations I’ve seen handle this phase well run a proper post-incident review within a fortnight, with the people who were on the calls, without blame, and turn the findings into a short list of funded changes. The ones that handle it badly commission a report from a large consultancy, receive it in four months, and file it. The Privacy Act now carries penalties that make the second approach a genuine financial risk rather than a reputational one, and the OAIC has shown it’s prepared to use them.

What separates the good outcomes #

Preparation, obviously, and by that I mean four specific things rather than a policy document. A tested backup that’s genuinely offline. A log retention window long enough to see a slow intrusion, which is 12 months as a floor. Phishing-resistant multi-factor authentication on every administrative account, which the Australian Signals Directorate’s Essential Eight has required at its higher maturity levels for years now and which stops the majority of the intrusions I’ve worked. And a contact list, printed, of who to ring, because the incident that takes out your identity provider takes out your email and your wiki with it.

The organisations that come out of a breach with their reputation intact tend to share one trait, which is that they told the truth early, in plain language, and updated it as they learned more. It’s not a communications strategy. It’s just what the OAIC, the customers and the journalists respond to, and it’s a lot easier to do if you were never tempted to say “no data was affected” in hour two.

The measured opinion, since you’ll want one #

I think the industry’s obsession with attribution, the question of who did it, is mostly a distraction during the first 72 hours. It matters to the police and to the insurer and to the board’s sense of being wronged. It doesn’t change a single containment action. Work out what they touched. Work out who you have to tell. Work out who did it later, if at all.

I proved a loaf on Saturday night before the phone rang, in the incident I keep describing, and it was over-proved and flat by the time I got back to it. I’ve since learned to bake on Fridays. Small lesson; you loose fewer loaves that way.

— Tom Abernathy, Security Editor, Technology Spectator

Back to top button