Skip to main content
Back
IntrusionAWS
9 min read

ARTEX agentic framework mounts EBS snapshots via a compromised IAM user

Victim
Unnamed organizations on AWS in Mandiant incident response engagements; the report does not say how many
When
since late August 2026Disclosed Oct 30, 2026
Entry
Compromised credentials of an AWS IAM user, supplied to an agentic offensive framework that ran the later AWS stages; how the credentials were obtained is not stated.

Google Threat Intelligence Group and Mandiant report that, since late August 2026, threat actors have operated the agentic offensive framework ARTEX in intrusions they responded to. In AWS, the framework was given the compromised credentials of an IAM user: it probed what the user could do with dry-run calls, tried to launch instances with potentially high-value roles, set up a persistent jump host with a public address, and launched keyless instances whose boot scripts mounted multiple EBS snapshots and volumes. The report calls the volume and disk growth likely preparation for exfiltration and does not confirm that data was read or taken.

What happened

Operators
Agentic framework
Internet
AWS IAM and APIIAM user credentials
Compromised AWS Account
EBS Snapshots and Volumes
Compromised IAM User
High-value Roles
Persistent Jump Host
Keyless Instances
EBS Snapshots
New 100 GB Volume

Replaying ARTEX in a compromised AWS account

  1. 1

    Compromised IAM user

    Since late August 2026; day not stated

    The operators give the framework the credentials of an AWS IAM user, and it signs its AWS calls with them; how the credentials were obtained, and the network the calls came from, are not stated.

  2. 2

    Permission probing

    Since late August 2026; day not stated

    Dry-run calls with placeholder identifiers test what the user can do on the EC2 compute, EBS volume and snapshot APIs; in at least one operation 27 events in 23 seconds probe permissions across security groups.

  3. 3

    Role harvesting attempt

    Since late August 2026; day not stated

    The framework launches instances with potentially high-value IAM roles attached, to fetch the roles' temporary credentials; the report says the attempts seemed to fail on invalid instance profile ARNs.

  4. 4

    Persistent jump host

    Since late August 2026; day not stated

    The framework creates a new SSH key pair and a public Elastic IP, is refused a custom security group, and launches persistent small instances on existing security groups as an externally addressable jump host.

  5. 5

    Snapshot harvesting

    Since late August 2026; day not stated

    Keyless instances mount multiple EBS snapshots and volumes through boot scripts, one with 17 snapshots, and a 100 GB volume and larger root disks are created, likely for exfiltration; whether data was taken is not confirmed.

Impact

The framework probed the IAM user's EC2, EBS volume and snapshot permissions, attempted to obtain role credentials through instance profiles, built a persistent jump host, and launched keyless instances that mounted multiple EBS snapshots and volumes, with volume and disk growth likely intended for exfiltration; whether data was read or exfiltrated is not confirmed.

Outcome

Where cloud-provider controls fit

This page reviews the playbook of one tool across unnamed victims, so it judges each control as deployed on your own AWS accounts: a row says what would happen if you have that control, and its caveat names the conditions. On those terms two rows would have stopped a step outright. A service control policy that denies ec2:RunInstances to all but approved administrators refuses the launch of the jump host, and one that also denies volume creation and attachment refuses the keyless instances and the new 100 GB volume. Both hold if the account is a member account of your organization and the compromised user is not one of your approved administrators. The other rows bound a step. A service control policy that refuses IAM user calls without MFA stays contained because the report does not say which credential signed the calls. The dry-run probes are still answered, and the report says the launches with high-value roles seemed to fail. An EC2 declarative policy that sets VPC Block Public Access would have blocked the jump host's internet-gateway path, subject to VPC or subnet exclusions, and restricting effective KMS permissions across the key policy, delegated IAM policies and grants would have kept an unauthorized user from using snapshots encrypted with that customer managed key. You would still need to find out how the credentials were obtained, which the report does not state.

Could this have been prevented?

In hindsight, not as a whole. 2 controls would have refused a step outright, and 5 controls would have limited one. 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.

2
Prevented
5
Contained
0
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.

Operators
Agentic framework
Internet
AWS IAM and APIIAM user credentials
Compromised AWS Account
EBS Snapshots and Volumes
Compromised IAM User
High-value Roles
Persistent Jump Host
Keyless Instances
EBS Snapshots
New 100 GB Volume

Replaying ARTEX in a compromised AWS account

1 · Compromised IAM user

1 Contained3 on other clouds

2 · Permission probing

1 Contained

3 · Role harvesting attempt

1 Contained3 on other clouds

4 · Persistent jump host

1 Prevented1 Contained3 on other clouds

5 · Snapshot harvesting

1 Prevented1 Contained3 on other clouds

Key facts

  • Google Threat Intelligence Group and Mandiant observed threat actors operating ARTEX and similar platforms in Mandiant incident response engagements from late August 2026, and published their report on October 30, 2026
  • In several cases the operators supplied the framework with compromised credentials and instructed it to run the later attack stages; how the credentials were obtained is not stated
  • In at least one operation, 27 events in 23 seconds of active burst time probed AWS permissions across different security groups
  • Instance launches with potentially high-value IAM roles seemed to fail on invalid instance profile ARNs, and the creation of a custom security group failed on permission issues
  • One instance mounting 17 snapshots was deployed in about 10 seconds, and a new 100 GB volume was created and root disks expanded to 60 GB, steps the report calls likely intended for exfiltration
  • The report names no victim, gives no calendar date for the AWS actions and does not say whether they come from one intrusion or several

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. 1.
    ARTEX and Friends: Enter Agentic Offensive FrameworksGoogle Threat Intelligence Group and Mandiant · Oct 30, 2026 · Researchframework description, compromised IAM user, permission probing with dry-run calls, failed instance profile escalation, denied security group creation, key pair and Elastic IP, persistent jump host, keyless instances mounting EBS snapshots and volumes, volume and root disk growth, burst timings, AWS guidance

The estate in the replay is illustrative and drawn at the level of attack stages; the victim's real environment is not public. Verdicts describe what a control does when it is in place before the incident. They are not an assessment of the victim's security program.

More cloud attack breakdowns

All breakdowns