SMSMC Blog

Ransomware that never touched a server: leaked AWS keys and S3 encryption

Attackers used leaked AWS keys to re-encrypt S3 data with keys only they held. What happened, what AWS changed in April 2026, and the access-key habits that matter more.

By Jim Venuto

The attack that used AWS’s own feature

In January 2025, researchers at Halcyon described a group they called Codefinger. It did not break into anyone’s servers or install malware. It called Amazon S3 using AWS access keys that were publicly exposed or otherwise compromised, keys that already had permission to read and write the victims’ data.

With those keys, the group rewrote the victims’ S3 objects using SSE-C, an S3 option where the customer supplies the encryption key with each request. The attackers supplied their own key. AWS does not store SSE-C keys, so once the objects were rewritten, nobody but the attackers could decrypt them. The group then set an S3 lifecycle rule to delete the files within seven days and left a ransom note in each affected directory. Halcyon said it had seen at least two victims and called it the first case it knew of that used SSE-C this way.

AWS told reporters that whenever it is aware of exposed keys, it notifies the affected customers.

What AWS changed in April 2026

In November 2025, AWS announced that S3 would disable SSE-C by default. The rollout began on April 6, 2026, across 37 AWS Regions:

  • New general purpose buckets have SSE-C disabled.
  • Existing buckets in accounts with no SSE-C encrypted objects also stop accepting new SSE-C writes.
  • Accounts that already use SSE-C keep their existing bucket settings unchanged.

A bucket owner who genuinely needs SSE-C can turn it back on per bucket, which requires the s3:PutEncryptionConfiguration permission. AWS’s stated reason is that SSE-KMS now gives the same protection with more flexibility, so SSE-C offers little practical benefit for most workloads.

That closes the specific trick Codefinger used for most small accounts. It does not close the door Codefinger walked through.

The real problem is the key, not the encryption option

Every step of this attack was an ordinary, permitted API call made with a valid key. The same key could just as easily have copied the data out, deleted it outright, or been used to change bucket settings. A small business with AWS usually has these keys in places nobody tracks:

  • A developer’s laptop, in a credentials file created years ago
  • A backup tool, website plugin or accounting integration that was set up once with an IAM user key
  • A code repository, a shared drive or a ticket where someone pasted a key to get something working
  • A former contractor’s machine

The lasting fix is to need fewer long-term keys. People can sign in through single sign-on and get short-lived credentials, and workloads running on AWS can use roles instead of a stored key.

Pointers that reduce the damage a leaked key can do

These are the reasonable first moves. They are not the whole job, and the order depends on what your account looks like today.

  1. Find every long-term access key. IAM’s access-last-used information shows when each key was last used. Keys that have not been used in months are the easiest to disable.
  2. Give each remaining key one job. A backup tool’s key should be able to write to the backup bucket and nothing else. A key that can read and rewrite every bucket is the key Codefinger needed.
  3. Separate backups from the account they protect. Locked copies in a different AWS account survive a key that can delete or rewrite production data.
  4. Watch for the unusual call. CloudTrail records S3 lifecycle changes and bucket encryption changes. An alert on a new lifecycle rule that deletes data, or on a bucket’s encryption setting being changed, is cheap and specific.
  5. Confirm the April 2026 default applies to you. If your account had any SSE-C objects before the change, your existing buckets still accept SSE-C. Decide whether you need it, and if not, block it in the bucket’s default encryption settings.

Who does what

The owner or executive sponsor decides which systems are allowed to hold AWS credentials at all, and accepts or rejects the exceptions. Whoever runs your AWS account, in-house or an MSP, inventories and rotates keys, scopes permissions, sets up separate backup storage and the alerts. Developers and vendors who integrate with AWS move to roles or short-lived credentials and tell you where any remaining key lives.

Where outside help fits

Most small AWS accounts grew one project at a time, so nobody has a full picture of which keys exist, what each one can do, and whether the backups would survive one of them leaking. Security Medic’s Cloud Security Assessment looks at that picture across AWS, Azure and Google Cloud and sets the order of work. CyberIntelPro leads the architecture side, such as how backups, accounts and permissions should be separated, and gives your team or MSP a clear list of what to change. Hudson Valley CISO keeps it governed: who owns each exception, and how often it is reviewed.

Want to talk it through? Book a free 30-minute call with Security Medic Consulting.

Sources