Incident response plan for a small team
One page of decisions, tested console access, backups the server cannot delete and short drills. What a team of one to five should prepare before an incident.
Incident response plan for a small team: the short answer
If your server is compromised right now, you need the response steps, not an incident response plan. Stop here and follow what to do if your VPS is hacked instead.
For a team of one to five people, an incident response plan is one page plus a few tested things. The page names who decides and who you call. It records how you get into the provider panel and the console when the server is no longer yours. It relies on backups you have actually restored. And you rehearse it in a short drill several times a year, so the first time you run it is not the night you need it.
This guide is for the weeks when nothing is wrong. It covers what to write down, what access to arrange, what to test and how often to drill, so a tired person can do the response steps at 3am.
Readiness is not response
The response guide asks you to do four things fast. Cut the network at the provider. Reach backups the attacker could not touch. Get into a console when SSH is gone. Reach the right people. You cannot improvise any of these on the night. Each one depends on something you set up earlier: a working panel login, a backup the server cannot delete, a console you have tested, and a call list someone wrote down.
If you want an official reference, use the current one. NIST (the US National Institute of Standards and Technology) published SP 800-61 Revision 3 in April 2025. It replaced Revision 2 from 2012, which NIST withdrew on 3 April 2025. Revision 3 is written as a Community Profile of the NIST Cybersecurity Framework (CSF) 2.0. It puts preparation under the CSF functions Govern, Identify and Protect, and it treats them as wider risk management that supports incident response, not as a step inside the response itself. That is the right way for a small team to think about it. Preparation is ordinary work you do every month, not the first page of an emergency.
The one-page incident response plan template
Copy this block into a text file, fill it in, and print it. It is meant to fit on one printed page. If yours grows past one page, move detail into the credential inventory further down and keep only pointers here.
INCIDENT RESPONSE PLAN: <team or project name>
Last reviewed: <date> Last drill: <date>
Printed copy kept at: <place> Off-site copy: <place NOT on our own servers or domain>
1. WHO DECIDES
Incident lead: <name> <phone>
Stand-in if lead unreachable: <name> <phone>
The lead (or the stand-in) may take every decision in section 2 alone.
2. DECIDED IN ADVANCE
Cut the network at the provider when: <unknown login / unknown SSH key / ransom note / abuse notice>
Default for a compromised server: REBUILD from a known-good backup. We do not clean it.
Ransom or extortion demands are decided by: <name>, after advice from: <lawyer / insurer>
Nobody pays or replies to an attacker before that person decides.
3. CALL LIST, IN THIS ORDER
1. VPS provider support: <URL or phone> account ID: <id>
2. VPS provider abuse: <address>
3. Domain registrar: <company> account: <id>
4. DNS host: <company> account: <id>
5. Email provider: <company> account: <id>
6. Customers or users: <who tells them, through which channel>
7. Must be told (law or contract): <regulator / client / insurer> deadline: <hours or days>
4. ACCESS WHEN THE SERVER IS NOT OURS
People with a provider panel login: <names> 2FA on each: <yes / no>
Recovery codes stored at: <offline place> last checked: <date>
Web console or rescue mode: <how to open it> last tested: <date>
Console password for root or admin user kept at: <place>
Spare admin SSH key kept at: <offline place> trusted clean device: <which>
5. BACKUPS
Repository: <where> Can the server delete it? <no / YES: fix this>
Last full restore test: <date> took: <minutes>
Repository password held by: <names>
6. OUT-OF-BAND CHANNEL
If our server or our email is the incident, we talk on: <phone group / address not on our domain>
7. AFTER
Notes kept in: <place> Review meeting within: <days> of the incidentHere is why each part is there.
Who decides, and the stand-in. In a small team the person who knows the server best is often the one on holiday. Name a stand-in, and give both of them the right to act alone. A plan that needs two people to agree will wait for the second person.
Decisions taken in advance. At 3am nobody should be debating whether to cut the network or whether to try cleaning the box. The response guide explains why you isolate at the provider and why you rebuild. Your plan only has to record that you agreed to do both. The ransom line names one person, because a demand that reaches several people by email can get several different replies.
The call list, in order. Your provider comes first, because the provider controls the network and the console. The registrar and the DNS host come next, because an attacker who gets into those accounts can move your domain or your mail. The last line is anyone with a legal or contract duty to be told. If a contract or a regulator gives you a deadline, write it down. Check that with your own lawyer or contract, not with this guide.
The out-of-band channel. If your email runs on your own domain and the incident is your DNS or your mail server, your normal email may not reach anyone. Agree on a phone group or an address on a domain you do not run, and test that every person is in it.
Where the plan lives. Keep one printed copy and one copy outside your own infrastructure. A plan stored only on the compromised server is no plan.
Technical preparation you can check today
Each item below ends with a check. Do the check, then write the date into the plan.
Backups you have actually restored
A backup you have never restored is a guess about a backup. Start with the difference between a provider snapshot and a real backup in VPS snapshots vs backups. A snapshot stored in the same account as the server is lost when that account is lost. Then keep at least one copy somewhere else, as in using a second VPS as an off-site backup target. If you have not picked a tool yet, encrypted restic backups on a VPS walks through one from install to restore.
Check today: restore one whole service to a scratch server, not one file. Time it. Write the time and the date into section 5 of the plan.
Backups the server cannot delete
An attacker with root on your server has every credential that server holds. If one of those credentials can delete the backup repository, the attacker can delete your restore points before they encrypt anything. There are two ways around this.
The first is an append-only repository. The restic REST server has an append-only mode, started with rest-server --append-only. Its README says this mode allows creating new backups but prevents deleting or changing existing ones. Restic's own docs add two cautions. The forget and prune commands need full access, so run them from a separate, well-secured machine and never from the server being backed up. An attacker can also add fake snapshots dated just after the real ones, so a later policy-based forget could remove the real ones. The docs suggest --keep-within as a mitigation, and it only protects you for the length of time you give it.
Borg has the same idea. In the Borg 1.4 docs you set it on the repository with borg config /path/to/repo append_only 1, or on the server side by putting borg serve --append-only in the command= of the client's key in authorized_keys. A few details matter. Running borg init through an append-only server does not create an append-only repository. Append-only mode frees no space, and borg compact is skipped. The next time anyone writes to the repository in normal mode, objects deleted in the meantime are removed for good, so a nightly prune job in normal mode undoes the protection. Append-only mode also limits only what Borg does. It does not stop a person with a shell on the backup host from deleting files.
The second way is a pull-based backup. The backup host logs in to the server and copies the data out. The server then holds no credential for the backup host at all, so there is nothing on it to steal.
Check today: on the production server, try to delete an old snapshot using only the credentials stored on that server. It should fail. If it works, an attacker can do it too.
The provider panel is part of your perimeter
The panel can reinstall the server and delete its snapshots. Whoever holds a panel login can undo every backup you keep inside that account. Turn on two-factor authentication (2FA) for every login, if your provider offers it. Store the recovery codes offline, in the place section 4 names. Keep a list of everyone who can log in, contractors included. For a clear split of what your provider handles and what stays your job, read what you take on when you rent a VPS. If you are unsure which side of that split you bought, managed vs unmanaged VPS spells it out.
Check today: log out, then log back in with your second factor. Find the recovery codes without help. Compare the panel's user list with your own list.
Console and rescue access, tested now
When SSH is gone, the provider's web console is how you reach the box. The response guide notes that a server with key-only login may need a root password reset before the console is any use. A reset during an incident costs time you may not have. So open the console today and log in through it. Find out from your provider's own docs whether it has a rescue mode and how you start it, then write the steps into section 4.
Check today: log in through the web console as a real user. Record the date.
A spare clean admin path
If the key on your laptop is the one that leaked, you cannot use that laptop to fix things. Keep a second admin SSH key that lives offline and is only used for emergencies. Name a device you trust that is not your everyday machine. The basics of creating and removing keys are in SSH key management basics.
Check today: log in with the spare key from the spare device, then put the key away again.
A credential inventory
The response guide tells you to rotate every credential the machine could see. Without a list, that step becomes a search through config files at 3am. Write the list now. One line per secret is enough.
SECRET WHERE ON THE SERVER ROTATE AT OWNER
Postgres app password /srv/app/.env database + app config <name>
SMTP relay login /etc/postfix/sasl_passwd mail provider dashboard <name>
Object storage access key /root/.config/rclone storage provider panel <name>
Backup repository password /root/.restic-pass backup tool key commands <name>
Deploy key for git repo /home/deploy/.ssh/id_ed25519 git host settings <name>
Payment or API tokens /srv/app/.env vendor dashboards <name>The paths are examples. Yours will differ, and what matters is that you know them. Keep the inventory with the plan, outside the server, and never write the secret values into it.
A rebuild path you have run once
Your default is to rebuild, so the rebuild has to be known work. Write the steps from an empty server to a working service. Start from the first ten minutes on a new VPS for the base, then add your own services. If they run in Docker Compose, backing up and upgrading a Docker Compose stack covers the parts that make a restore repeatable.
Check today: run the steps once on a scratch server, with the plan open. Note every step that was missing from the page.
One account per person, with an end date
CrowdStrike describes FAMOUS CHOLLIMA as a DPRK-nexus adversary whose operatives pose as legitimate remote workers, in some cases hired through IT staffing firms. For a small team the lesson is simple. Every person, contractor included, gets their own account and their own key, and the plan says who removes them on their last day.
Why the clock matters: what Qilin shows a small team
Black Kite's 2026 Ransomware Report covers April 2025 to March 2026. It counts 7,551 publicly disclosed ransomware victims, found by watching ransomware groups' leak sites and checked by Black Kite's research group. Attacks that never reach a leak site are not in that count, so the true number of attacks is higher. The top group was Qilin, with 1,358 claimed victims, which Black Kite says is nearly twice its nearest rival. On its actor-models page Black Kite says Qilin went from 250 victims in the previous period to 1,358, a 443% increase. By our own arithmetic, 1,358 out of 7,551 is about 18%, roughly one victim in five or six.
The activity did not stop in March. ZeroFox reported at least 165 incidents for Qilin in a single month of 2026, a record for one group so far that year. In a post dated 28 October 2026, the Acronis Threat Research Unit wrote that Qilin had already claimed more than 900 victims in 2026.
How they get in matters more to you than how many they hit. Cisco Talos wrote on 2 April 2026 about ransomware in Japan in 2025 and the Qilin cases it investigated. Talos says Qilin primarily relies on stolen credentials for initial access, obtained through Telegram, Breach Forums and other platforms. Across those cases, Talos found that ransomware ran on average about six days after the first compromise. That is an average over Talos's own cases, not a rule. It still shows the shape of the risk: there is often a gap between someone getting in and the damage.
On a VPS, the stolen credentials that matter are your SSH keys, your provider panel login, your backup credentials and the logins of the apps you run. A gap of days is time a team with detection habits can use. These habits fit a team of one to five.
- Once a week, look at who logged in and what they ran. Auditing user commands on your server shows how to record that, so there is something to look at.
- Notice new SSH keys. A key in
authorized_keysthat nobody on the team added is a reason to open the plan. Keeping a list of which keys belong to whom is what makes a new key stand out. - Notice new user accounts, and any account that should have been removed when someone left.
- Know what normal looks like on your box: which services listen, which users exist, and roughly how busy the CPU and disk are. You cannot spot something strange without a picture of normal.
A check that runs once a week has a real chance of landing inside a gap of a few days. A check that runs once a year almost never will.
Drills a tiny team can actually run
A drill shows you which parts of the plan only work on paper. Keep drills short, so you actually run them.
A 45-minute tabletop
One person reads a scenario out loud. Everyone else says what they would do, using only the printed plan. Nobody touches a real server. Give each scenario about fifteen minutes.
- A ransom note on a Monday morning. The web root is encrypted and a text file asks for payment. Who decides? Who cuts the network at the provider, and do they have a panel login? Where is the last backup from before the attacker got in, and could the server reach it? Who tells customers, and on which channel?
- A leaked SSH key. A contractor's laptop was stolen, or a key turned up in a public repository. Which servers trust that key? Who removes it, and how fast? According to the credential inventory, what else could that key reach?
- A stolen panel login. Someone else has logged in to your provider account. Can you still log in? Where are the recovery codes? If the attacker deleted your snapshots, where is your off-site copy? How does provider support check that you are the real owner?
Every "we would have to look that up" is a gap. Write it down.
A timed restore rehearsal
Once a tabletop feels easy, do the real thing on a scratch server. Measure two numbers. Time-to-isolate runs from "we decide to cut the network" to the scratch server being cut off at the provider. Time-to-restore runs from an empty server to a working service with real data. Write both numbers into the plan, and set your own targets. The point is that the real night is not your first measurement.
Write down what broke
Every drill finds something. A recovery code was never printed. A teammate is missing from the phone group. The console password in the plan is out of date. A restore step needs a package nobody wrote down. Keep a short list. Fix each item before the next drill, and note the fix in the plan.
How often to rehearse the plan
Short and repeatable beats annual and elaborate. People join and leave. Providers change their panels. Keys get copied to new laptops. You add a service with new secrets. A plan written a year ago describes a team and a server that no longer exist, and a plan nobody has run usually fails the first time it is needed. A 45-minute tabletop and a restore test each quarter is a sensible start for a small team. Pick a rhythm you will keep.
Some changes should trigger a review straight away, without waiting for the next drill.
- Someone joins or leaves. Add or remove their panel login, their SSH key and their place in the phone group. Check who holds the recovery codes and the backup password.
- You change provider. The panel, the console, the rescue mode, the abuse contact and the way snapshots work all change with it. Test console access again from the start.
- You add a service. Add its secrets to the credential inventory and its data to the backups. Then restore it once, because a backup set you changed is a backup set you have not tested.
- A key or a device is lost. Remove the key from every server that trusts it, then check the credential inventory for anything else that device could reach.
When it happens anyway
If you see a ransom note, an unknown key or an abuse notice from your host, the runbook is what to do if your VPS is hacked. This plan is what makes each of its steps possible. Section 2 already says you will cut the network at the provider and rebuild. Section 4 gets you into the console. Section 5 tells you where the clean backups are. The credential inventory turns rotation into a checklist. The call list tells you who to phone, in order. Print the plan today, and put the date of your first drill in the calendar.
FAQ
Do I need an incident response plan if I am the only admin?
Yes, and it matters more. A solo admin has no colleague to remember the panel login or the backup location while they are asleep or away. Write the one-page plan anyway. Name a stand-in who can at least call the provider and cut the network, and make sure that person can find the printed copy and the recovery codes.
How is an incident response plan different from a backup plan?
A backup plan answers where your data is copied and how you restore it. An incident response plan answers who decides, who you call, how you get in when the server is not yours, and which backups you trust after an attack. Backups are one section of the plan. The plan also has to make sure an attacker on the server cannot delete them, for example with an append-only repository or a pull-based backup.
How often should a small team test its incident response plan?
Several times a year, with short drills instead of one long annual exercise. A 45-minute tabletop and a timed restore to a scratch server each quarter is a sensible start. Also review the plan when someone joins or leaves, when you change provider, when you add a service, and when a key or a device is lost. Each of those changes the accounts and backups the plan depends on.
Where should I keep the incident response plan?
Keep one printed copy where the incident lead and the stand-in can reach it. Keep a second copy outside your own infrastructure, such as a shared drive on a domain you do not run. Never keep the only copy on the server or the email system the plan is meant to protect. Store the recovery codes offline, apart from the plan, and record in the plan only where they are.