The First 72 Hours After a Data Breach
What to do in the first three days of a breach: containment that preserves evidence, scoping, notification timing in general terms, and communications.

The call usually comes in a form that is not obviously a breach. A customer says they received an invoice from you with the wrong bank details. A developer notices a database export in a storage bucket nobody uses. An employee mentions that their password stopped working on Tuesday and they had to reset it twice. Somebody finds a file share full of files with a new extension.
What happens in the next three days determines most of the eventual cost, and most of the expensive mistakes are made in the first three hours by capable people acting sensibly without coordination. Someone reimages the laptop that held the only evidence. Someone emails the whole company from the compromised mail system. Someone resets one password while the attacker still holds a valid session token. This is a sequence designed to avoid that.
Hour zero: declare it and take control
Before any technical action, do three things that cost ten minutes.
Name an incident lead. One person who makes decisions and whose job is coordination rather than analysis. Without this you get parallel work and contradictory actions. If your organization has an incident response plan, this is who it names, which is the main reason to have written one.
Open a single log. One document, one place, with every entry timestamped in a stated timezone, ideally UTC. Record what was observed, who did what and when, what was decided and by whom, and what was not done and why. This log does three jobs: it keeps the team aligned, it becomes the basis of your regulatory account, and it protects individuals from later second-guessing. Keep it outside the systems that may be compromised.
Move communications out of band. If email, identity or chat might be in scope, assume the attacker is reading. Agree a channel on separate identities and tell everyone involved to use it. Discussing the investigation of a mailbox compromise inside that mailbox system is a mistake that gets made surprisingly often.
Hours 0 to 2: contain without destroying the evidence
Containment and forensics pull in opposite directions, and the resolution is nearly always the same: isolate, do not erase.
What to do:
- Disconnect affected hosts from the network while leaving them powered on. Pull the cable or disable the switch port or network adapter. If you have the capability to capture memory from a live host, do it before isolating, because volatile evidence disappears at shutdown.
- Disable compromised accounts and revoke their sessions and refresh tokens. A password reset alone leaves active sessions working, which is the single most common containment failure in cloud environments. Revoke, then reset, then re-register MFA.
- Rotate credentials the attacker may hold, prioritizing anything with wide reach: administrative accounts, service accounts, API keys, keys in code repositories, remote access credentials, and any shared password.
- Close the entry point if you know it, such as blocking an exposed service, applying a patch to an internet-facing appliance, or removing a malicious mailbox rule and the forwarding configuration behind it.
- Remove exposed data from public reach if the incident is an exposure, then check for cached copies and ask search engines to remove them. The FTC's business guide on breach response is specific about this step.
What not to do: do not reimage, do not run cleanup tools across affected machines, do not delete the mailbox rules before recording them, do not restore from backup over the top of evidence, and do not log into compromised systems with your most privileged credentials, because that is how an attacker collects a better one.
Hours 1 to 6: preserve what you will lose first
Evidence does not expire evenly. Logs disappear on a rotation schedule that nobody checks until it matters, and that is usually the thing you need most.
- Extend log retention immediately wherever you can change a setting: cloud identity sign-in and audit logs, email audit logs, EDR telemetry, firewall and VPN logs, web proxy logs, DNS query logs, and application logs.
- Export rather than trust. Pull copies to a separate location now, even if you cannot analyze them yet. Exporting takes minutes and the window may be days.
- Record the volatile things, including currently active sessions, scheduled tasks, running processes, recently created accounts, and the configuration state you are about to change.
- Take disk images of key hosts before rebuilding them, or at minimum snapshot virtual machines. Note who took each image, when, and where it is stored, since a simple chain-of-custody note is often enough and its absence is not.
- Preserve the originating artifact, the phishing email with full headers, the malicious file, the ransom note, the fraudulent invoice. Keep originals rather than screenshots.
- Write down what you cannot preserve and why. Gaps that are documented are defensible, while gaps discovered later look like concealment.
If the incident involves ransomware, the CISA StopRansomware guide and the NIST guidance on detecting and responding to destructive events both set out the preservation sequence in more detail, and our guide to practical ransomware defense covers the decisions specific to that case.
Hours 2 to 24: establish scope
Scope is the question every other decision depends on, including whether you have to notify anyone. Work it in this order.
Build an identity timeline. For each suspected account, list every sign-in with time, address, device and application, the point where behavior changes, and everything the account did after that point. In cloud environments this is the backbone of the investigation, because the attacker's actions are mostly authenticated actions.
Determine what data was reachable, then what was actually touched. These are different findings and the difference is the whole notification question. Reachable is what the compromised access permitted. Touched is what logs show being read, downloaded, exported or sent. Say which one you are reporting, every time.
Look for lateral movement and persistence. New accounts, new administrative role assignments, new application registrations and consent grants, new mailbox rules and delegations, scheduled tasks, remote access tools, VPN profiles, and changes to MFA registrations. Assume the attacker expected to be evicted and planned to come back.
Characterize the data. Categories of personal data, whether special categories or financial data are involved, number of individuals affected and in which countries, and whether anything was encrypted or tokenized in a way that reduces the impact. Counting carefully now avoids revising a public number later.
Check for the second incident. Credential theft frequently precedes everything else, and a payment diversion attempt often runs in parallel with a technical intrusion. Our explainer on what counts as a data breach is a useful reference for deciding what you are actually dealing with.
Day one: get the obligations clock started
This is a legal question, and the point of raising it on day one is not to answer it yourself but to make sure somebody competent is answering it in parallel with the technical work.
In general terms, the obligations that commonly apply:
- Data protection law. Many regimes require notification to a supervisory authority within a short, fixed period measured from the point you become aware of a personal data breach, with 72 hours being a widely used figure, and separate duties to inform affected individuals when the risk to them is high. The European Data Protection Board's guidelines on breach notification examples are a good illustration of how regulators reason about these judgments, including partial notification where facts are still emerging.
- Sectoral rules, which can be considerably faster and may apply to financial services, healthcare, critical infrastructure, telecommunications and listed companies.
- Contractual duties. Customer and supplier contracts frequently contain notification windows shorter than the law requires. If you process data on behalf of others, you probably owe them notice quickly, and your contract will say how.
- Insurance. Policies typically require prompt notification and the use of approved responders. Call the hotline before you engage anyone.
- Law enforcement, which does not usually change your other obligations but matters for fraud recovery and attribution. In the US that includes CISA for incident reporting and the FBI's Internet Crime Complaint Center for cyber-enabled fraud.
Two practical notes. Engage counsel early, because the sequence and wording of what you record and send has consequences, and privilege rules vary. And decide explicitly when you became "aware", with reasoning in your log, since that timestamp is the one a regulator will test. Our article on breach notification covers the recipient's side of this.
Hours 24 to 72: eradicate and recover in the right order
Recovery that runs ahead of eradication produces a second incident. The sequence that works:
- Confirm the entry point and prove it is closed. If you cannot identify it, treat the whole environment as suspect rather than declaring victory on the systems you happened to look at.
- Rotate credentials environment-wide, rather than just the obviously affected ones. Include service accounts, API keys, certificates where feasible, and any directory-level secrets. Where an identity platform may have been compromised, consider that every token in it is suspect.
- Rebuild rather than clean compromised hosts, from known-good images, patched before they rejoin the network.
- Restore data from a backup predating the compromise, verified and scanned, and check for malicious persistence inside the backup itself.
- Reconnect in stages, with monitoring raised and specific detections in place for the attacker's observed techniques.
- Watch for re-entry for weeks, not days. This is when the attacker tries the credential you missed.
Resist the pressure to call it finished when systems come back. NIST's incident response guidance frames recovery as part of a continuing risk management cycle rather than an end state, and that framing is worth borrowing when an executive asks whether it is over.
Communications: say less, earlier, and accurately
Most reputational damage comes from the handling rather than the breach.
- Prepare a holding statement on day one, before you need it: what happened in one sentence, what you are doing, what you are not yet able to confirm, and when you will next update. Having it ready removes the choice between silence and speculation.
- Tell your own staff first, with a short script for anyone who speaks to customers. Employees learning about it from a customer or the press creates its own problem.
- Use one spokesperson. Consistency matters more than eloquence.
- Never state numbers you might revise. "We are still determining the scope" ages far better than a figure that triples.
- Be careful with "no evidence". It is true and it sounds like a denial, and it collapses badly if evidence appears. Say what you have checked and what you have not.
- Give people something to do when individuals are affected: what to watch for, what you will never ask them for, and how to reach you. Expect follow-on fraud using the breach as a pretext, which we cover in scams after a data breach.
- Publish updates on a channel you control and keep it current, since a stale status page implies you have lost interest.
Suspicious message during an incident? Paste the message or link into CyberWatch AI Scan for a free check before anyone responds to it.
The mistakes that recur
- Reimaging the first affected machine, which is usually the one that would have explained everything.
- Resetting passwords without revoking sessions and tokens.
- Coordinating the response inside the compromised email system.
- Letting log retention lapse during the first day because nobody owned that task.
- Containing one account and missing the mailbox rule, consent grant or remote access tool that restores the attacker's position.
- Announcing a number before scoping is finished.
- Calling the insurer after engaging a forensics firm.
- Treating "systems restored" as the end and skipping the review.
When it is genuinely over, run an honest review within two weeks while memories are fresh. The question is not who erred, but which specific thing you will change: a log you will now retain for longer, a control you will enforce, a contact you could not reach, a decision nobody had authority to make. Feed each one into the plan and into the rest of your program, as described in our small business cybersecurity guide. The organizations that handle a second incident well are the ones that wrote down what the first one taught them.
Frequently asked questions
What is the very first thing to do?
Write down the time and what you know, then decide who is leading. An incident with no named lead produces parallel, conflicting actions in the first hour, which is when the most damage is done. One person decides, one person records, everyone else reports to them.
Should we shut everything down immediately?
Isolate rather than destroy. Disconnecting a machine from the network stops the spread while keeping memory and disk intact; powering it off or reimaging it destroys evidence you will need and may trigger encryption in some ransomware cases. Isolate, then decide.
What evidence matters most and how long do we have?
Logs are the thing you lose first, because retention windows are often days. Extend or export logging before anything else, including cloud sign-in logs, email audit logs, firewall and VPN logs, and EDR telemetry. Memory from a live compromised host is valuable but perishable, so capture it before isolating if you have the capability.
When do we have to notify a regulator?
It depends on your jurisdiction, your sector and the type of data, and the answer should come from counsel rather than a blog. Many data protection regimes set a deadline measured from the moment you become aware, commonly 72 hours, and sectoral and contractual deadlines can be shorter. Start the clock question on day one rather than day three.
Do we have to tell customers straight away?
The legal duty and the right moment are different questions. Premature notification with wrong numbers damages trust more than a short, accurate delay, while silence after the facts are known is worse than either. Prepare a holding statement on day one so you are never choosing between speculation and no comment.
Should we call our insurer before or after bringing in help?
Before. Many cyber policies require prompt notification and require you to use approved responders, and engaging your own forensics firm first can affect coverage. The insurer's hotline is often also the fastest route to experienced responders.
Can we talk about the incident over email and chat?
Not if identity or email systems are in scope. Assume the attacker is reading. Move coordination to a pre-agreed out-of-band channel on separate identities, which is one of the things an incident response plan should establish in advance.
When is the incident over?
Not when systems are back. It is over when you have confirmed the entry point is closed, credentials and tokens across the environment have been rotated, monitoring has run clean for a defined period, and the notification and regulatory work is complete. Re-entry through the same door within weeks is a common pattern.
Sources
- SP 800-61 Rev. 3, Incident Response Recommendations and Considerations for Cybersecurity Risk Management, NIST
- Data Breach Response: A Guide for Business, US Federal Trade Commission
- Incident management, UK National Cyber Security Centre
- Incident management: Plan your cyber incident response processes, UK National Cyber Security Centre
- Guidelines 01/2021 on Examples regarding Personal Data Breach Notification, European Data Protection Board
- SP 1800-26, Data Integrity: Detecting and Responding to Ransomware and Other Destructive Events, NIST
- #StopRansomware Guide, CISA
- IRF Incident Reporting, CISA
- Home Page - Internet Crime Complaint Center (IC3), FBI Internet Crime Complaint Center


