Your cloud account isn't a safe by default
Ransomware in the cloud rarely looks like the version you've seen in the news. There's no ransom note on a desktop and no locked laptop. More often, someone gets hold of a credential, finds your storage, and either encrypts it, deletes it, or copies it out and threatens to publish it.
The cloud provider keeps the data center running. You're responsible for who can get into your account and what they can do once inside. That split is called the shared responsibility model, and most cloud ransomware incidents happen on the customer's side of the line.
How attacks actually happen
Attackers don't usually break the cloud. They log in to it.
The common entry points are boring:
- A stolen or reused password on an account with no multi-factor authentication.
- An access key committed to a public code repository, or left in a config file on a server.
- A phishing email that captures a session token, which lets the attacker skip the password and the MFA prompt.
- A forgotten admin account belonging to someone who left two years ago.
Once inside, the attacker looks around. They list your storage buckets, your databases, your virtual machines, and your backups. Then they pick a method.
Three ways your data gets held hostage
Encrypt in place. The attacker re-encrypts your files using a key only they control. In some storage services, they can use your own account's encryption features against you, setting a new key and deleting the old one.
Delete and demand. They wipe the data and the backups, then ask for payment. It's crude, and it works when nobody has an independent copy.
Steal and threaten. They copy the data out and threaten to leak it. Restoring from backup doesn't help here, because you still have a breach on your hands.
That third type is why backups alone aren't a complete answer. Keep it in mind when you plan.
Start with identity, not tools
Most of your protection comes from locking down who can sign in. This is where I'd spend the first week.
- Turn on MFA for every human account, and use hardware keys or authenticator apps for admins. SMS codes are better than nothing, but I wouldn't rely on them for anyone with broad permissions.
- Cut the number of admins. For a small or mid-sized company, two to four people with full administrative rights is a sensible range. If you have fifteen, you have fourteen too many.
- Replace long-lived access keys with short-lived credentials wherever your provider supports it. Where you can't, rotate keys every 90 days. That's a common rule of thumb, not a magic number, but it limits how long a leaked key stays useful.
- Remove unused accounts monthly. Check for anything that hasn't signed in for 60 to 90 days.
Pay particular attention to the root or owner account. Lock it away with MFA, and don't use it for daily work.
Backups that survive an attacker
Here's where many teams get a false sense of safety. If your backups live in the same account, with the same admin credentials, an attacker who gets in can delete them along with everything else.
A backup that the attacker can delete is just another copy of the target.
What you want is immutability: a copy that nobody can alter or erase for a set period, even with admin rights. Major providers offer this under names like object lock or immutable vaults. Set a retention window of 14 to 35 days as a starting point, based on how long it might take you to notice a problem. If attackers sit quietly for three weeks before striking, a 7-day window won't help.
Follow the long-standing 3-2-1 guideline: three copies, on two different types of storage, with one offsite. In the cloud, I'd translate "offsite" as "in a separate account, ideally with different credentials and a different administrator." A second region in the same account isn't enough.
My pick: a separate backup account with immutable storage. I wouldn't bother with elaborate cross-region replication of production data as a ransomware defense. Replication copies the damage just as quickly as it copies the good data.
Turn on versioning and logging
Versioning on storage buckets keeps older copies of each file when it's overwritten. If an attacker encrypts your files, the previous versions may still be there. Turn it on, and add a lifecycle rule to expire old versions after 30 to 90 days so storage costs stay sane. Extra storage for versioning typically adds a modest amount to your bill, though it depends heavily on how often your files change.
Logging matters just as much. You can't investigate what you didn't record. Enable your provider's audit log, which records who did what and when, and send it to a place the attacker can't edit. Keep at least 90 days; a year is better if compliance applies to you.
Then set alerts for the events that signal trouble:
- A new access key created for an admin account.
- MFA being disabled.
- Backup retention settings changed.
- A sudden spike in deletions or in data leaving your network.
Five or six well-chosen alerts beat fifty noisy ones that everyone ignores.
Limit the blast radius
Assume one account will eventually be compromised. Design so that it can't take everything with it.
Separate production, development, and backups into different accounts. Give each application only the permissions it needs, so a web server that reads files can't also delete them. This is called least privilege, and it feels tedious until the day it contains an incident to one system instead of twenty.
Practice the restore
A backup you've never restored is a hope, not a plan. Test a full restore of at least one important system every quarter. Time it. If your business can tolerate four hours of downtime and the restore takes two days, you've learned something valuable while it was cheap to learn.
Write down the first hour of your response, too: who gets called, who can disable accounts, who talks to your insurer and legal counsel. Keep a printed copy, because your email might be part of what's down.
Should you ever pay?
Law enforcement agencies in many countries advise against paying, and I agree. Payment doesn't guarantee you'll get working decryption tools, and it marks you as a business that pays. Some jurisdictions also restrict payments to sanctioned groups, so talk to legal counsel before anything else.
The best way not to face that choice is to make it unnecessary: strong sign-in controls, isolated immutable backups, and a restore you've actually tested.
A practical starting list
If you only do five things this month:
- Enforce MFA on all accounts.
- Shrink the admin list.
- Create a separate backup account with immutable storage.
- Turn on versioning and audit logging.
- Run one restore test and record the time.
None of it is exotic, and most of it costs more in attention than in money. That's the good news.