What the demonstration did
Replaying Aqua's Bucket Monopoly demonstration / reported to AWS February 2024; the runs are not dated
- 1
The service bucket is claimed first
Not dated · reported to AWS February 16, 2024In its own account, the attacker creates the S3 bucket under the name CloudFormation will use for the victim in a region the victim has not used, opens it to the public and grants the AWS service read, write and list on it.
- 2
Templates go to the attacker's bucket
Not dated · reported to AWS February 16, 2024The victim's CloudFormation use in that region writes the uploaded template into the attacker's bucket, and a function triggered on the write adds a role the attacker can assume before the stack deploys.
- 3
Code and roles land in the victim account
Not dated · reported to AWS February 16, 2024CloudFormation deploys the altered template and creates the injected role where the deploying identity holds IAM permissions, and the victim's Glue job runs a script altered in a claimed bucket with the job's role.
In the demonstrations, a claimed bucket gave the attacker the victim's CloudFormation templates and a role injected into the victim's stack, and code execution in the victim's Glue job; EMR, SageMaker, Service Catalog and CodeStar showed tampering, data leakage or denial of service. No exploitation in the wild is reported.
What Bucket Monopoly shows about your buckets
When a service writes to or reads from a bucket by a name that someone else can predict, the account that creates that name first receives your files; in Aqua's demonstrations that gave the bucket's owner the templates, a role in the stack and code in a Glue job. In your AWS organization, a service control policy that denies S3 reads and writes outside the organization bounds the write and the read, and one that requires a permissions boundary caps a role a stack creates; neither binds the management account. A resource control policy keeps principals outside the organization from assuming your roles, and became available in November 2024, after this research. AWS fixed the six services from March to June 2024.
Would the controls have stopped it?
In hindsight, no control would have stopped it. 4 controls would have limited how far it went. 2 steps have no cloud-provider control. The counts below are control assessments, not attack stages: several can address the same step. The same replay with the controls in place before the incident shows where each phase would have been refused or bounded.
- 0
- Prevented
- 4
- Contained
- 2
- Not covered
How it could have been prevented
The same replay with the controls in place. Green marks where a step is refused, blue where it is bounded, and red where no control reaches it. Open a row to read the control, why it acts, and its caveat.
Replaying Aqua's Bucket Monopoly demonstration / reported to AWS February 2024; the runs are not dated
1 · The service bucket is claimed first
1 Not covered3 on other clouds2 · Templates go to the attacker's bucket
1 Contained1 Not covered3 on other clouds3 · Code and roles land in the victim account
3 Contained6 on other cloudsKey facts
- Aqua Nautilus reported the CloudFormation, Glue, EMR, SageMaker and CodeStar issues to AWS on February 16, 2024, and the Service Catalog issue on February 18
- The 6 affected AWS services built bucket names from a service prefix, an account identifier and, in most patterns, the region; Aqua found the CloudFormation account hashes in public code repositories and issue trackers
- In the CloudFormation demonstration, the victim's template was written to the bucket in the researchers' own account, and a function triggered on writes added a role the attacker could assume before the stack deployed; the write-up does not state that the role was assumed
- In the Glue demonstration, the victim's job ran a script the researchers had altered in the claimed bucket, and the console showed a cached copy
- EMR, SageMaker Canvas, Service Catalog and CodeStar showed tampering, data leakage or denial of service
- AWS confirmed fixes from March 16 to June 26, 2024 that pick another name when the bucket exists, and stopped new CodeStar projects; AWS told The Register that no customer action is required, and no exploitation in the wild is reported
Sources
Every fact in this breakdown comes from the public reporting below. The replay is a model of that reporting, not a reconstruction of the environment it describes.
- 1.Bucket Monopoly: Breaching AWS Accounts Through Shadow ResourcesAqua Security · Aug 9, 2024 · Researchnaming pattern, bucket claim, bucket policy, CloudFormation template tampering, deployment identity, Glue script tampering, EMR, SageMaker Canvas, Service Catalog, CodeStar, disclosure timeline, AWS fixes, recommendations
- 2.AWS 'Bucket Monopoly' attacks could allow complete account takeoverThe Register · Aug 7, 2024 · NewsAWS statement, venue, affected services, hash discovery in public code
The estate in the replay follows the researchers' description of their own test environment. Verdicts describe what a control does when it is in place before the run. They are not an assessment of any company's security program.