AWS | Cloud Security | Qubole | Supply Chain | IAM | Data Exfiltration
Source Material: Wiz Research, Sentra Research, Qubole Documentation | Technology: AWS IAM, AWS STS, Amazon S3, Secrets Manager, SSM Parameter Store, and Lambda | Targeting: Opportunistic – Organizations with current or legacy Qubole AWS integrations
Executive Summary
Intelligence indicates that a threat actor using IAM credentials from an AWS account, which is associated with Qubole platform activity, assumed trusted integration roles in multiple customer AWS environments in October. This was active supply chain abuse of standing third-party trust, not exploitation of a published vulnerability. Deepwatch has observed an uptick in the abuse and compromise of third-party AWS access keys, as such even non-Qubole customers are advised to review third-party access and limit roles to the extent possible.
After assuming customer roles, the actor enumerated cloud resources, retrieved Lambda configuration and secrets from AWS Secrets Manager and Systems Manager Parameter Store, validated discovered AWS, Microsoft Entra ID, and Okta credentials with TruffleHog, and accessed Amazon S3 data. Dozens of Qubole customers were impacted. Dormant Qubole integration roles were especially significant because their trust policies still permitted access despite little or no legitimate vendor activity.
Organizations that currently use or previously used Qubole should inventory all AWS roles and access keys associated with the integration, remove unused trust, investigate the indicators and behaviors in this advisory, and rotate any secrets that may have been read. A valid External ID did not prevent the observed access. Public reporting does not establish how the Qubole-associated credentials were obtained, and no public Qubole incident statement explaining the initial compromise was identified at publication time.
Threat Overview and Strategic Impact
Qubole Data Service integrations can use cross-account IAM roles that customers create in their own AWS accounts. The trust policy permits a Qubole AWS account to call STS AssumeRole, typically with an External ID, and the resulting temporary credentials inherit the permissions attached to the customer role. In the observed campaign, valid credentials from the source account allowed the actor to use this intended trust path. Every role-assumption request attributed to the actor included a valid External ID and succeeded, demonstrating that an External ID is not a secret or sufficient control when a trusted vendor identity is compromised.
The actor used principal ID exclusively in the malicious activity observed by Wiz. A second principal was used for both malicious and legitimate Qubole traffic, so it requires network and behavioral context. Sessions often used the prefix quiam-, including quiam-enum and quiam-scope-full. Activity originated from Mullvad VPN and OVH infrastructure rather than the AWS infrastructure used for legitimate platform traffic.
Post-access behavior included broad List and Describe activity across S3, EC2, RDS, IAM, KMS, and data services; retrieval of Lambda environment variables, Secrets Manager values, and SSM parameters; and opportunistic S3 collection using ListBuckets, ListObjects, HeadObject, and GetObject. The actor also tested discovered credentials against AWS, Entra ID, and Okta. This creates potential exposure well beyond the original role because retrieved secrets may enable access to other cloud accounts, identity providers, SaaS platforms, or downstream systems.
Other observed customer-owned AWS access keys believed to support Qubole integrations being abused in another cluster of activity. That activity has not been linked to the vendor-account role-assumption campaign and should be investigated as a separate lead. No reliable attribution has been established; a public claim by an actor using the name “n0n” remains unverified.
Security Hardening and Recommendations
- First, address the Qubole-specific exposure: inventory current and legacy integration roles and access keys, remove unused trust, investigate the published indicators, and rotate secrets that may have been read. Coordinate any required trust-policy changes with the vendor to avoid disrupting legitimate services.
- More broadly, maintain a centralized inventory of third-party integrations and non-human identities with a documented owner, business purpose, approved permissions, data access, and expiration or review date. Prefer short-lived workload credentials over long-lived keys. Restrict trusted principals to the narrowest supported identity, bind access to documented source networks or private endpoints, use unique external IDs, enforce least privilege, segment sensitive data, and require security review before a vendor’s access scope expands. Revoke dormant integrations rather than leaving them monitored indefinitely.
- Maintain organization-wide cloud and identity logging, including object-level access for sensitive stores, and baseline each vendor’s expected networks, session patterns, services, and access volume. Contractually require prompt incident notification, published egress ranges, credential-rotation support, and security assurance evidence. Exercise response procedures that can rapidly suspend vendor access, invalidate sessions, rotate downstream secrets, preserve logs, determine affected data, and engage legal, privacy, procurement, and communications teams.
Detection Strategy
Use AWS CloudTrail to identify STS AssumeRole events into Qubole integration roles, especially when the source address is outside AWS infrastructure, the role was dormant, the session name begins with quiam-, or the request uses a listed principal or network indicator. Correlate the role assumption with rapid multi-service enumeration, unusual GetFunctionConfiguration, GetSecretValue, GetParameter, and GetParameters calls, and subsequent S3 ListObjects, HeadObject, or GetObject activity.
Detections should combine identity, network, session, and behavior. TruffleHog user-agent activity and bursts of GetCallerIdentity from newly discovered access keys are useful corroborating signals but can also occur during legitimate security testing.
How Deepwatch Protects Our Customers
For customers that provide relevant AWS CloudTrail, S3, KMS, endpoint, and identity-provider telemetry, Deepwatch MDR can correlate anomalous third-party role assumption with subsequent discovery, secret retrieval, credential validation, and cloud-data access. Analysts can prioritize high-signal combinations such as a dormant Qubole role activated from VPN or hosting infrastructure, quiam- session names, the malicious-only principal ID, rapid cross-service enumeration, and secret-access events.
Deepwatch can also support targeted hunting across AWS, Microsoft Entra ID, and Okta telemetry and help customers determine whether activity represents expected vendor operations or unauthorized use. Customers should provide an inventory of Qubole integration roles, approved AWS principals, expected source networks, and known legitimate session patterns to improve fidelity.
Detection coverage depends on the telemetry enabled and onboarded. CloudTrail management events establish role assumption and many enumeration actions, but S3 object reads require S3 data events or server access logs. Deepwatch cannot confirm customer exposure where the necessary logs were not retained or ingested.
Relevant Detections
Please visit Guardian Platform to access the relevant detections for this activity.
Technical Artifacts
Please visit Guardian Platform to access the associated technical artifacts.
Threat Object Mapping
Intrusion Set:
- Unattributed.
Attack Pattern (MITRE ATT&CK/MITRE ATLAS):
| Tactic | Technique | Technique ID | Associated Threat Activity |
| Initial Access / Defense Evasion | Valid Accounts: Cloud Accounts | T1078.004 | Used Qubole-associated credentials to assume trusted customer roles through an intended cross-account access path. |
| Discovery | Cloud Infrastructure Discovery | T1580 | Enumerated S3, EC2, RDS, IAM, KMS, clusters, databases, and other AWS resources. |
| Credential Access / Collection | Cloud Secrets Management Stores; Data from Cloud Storage | T1555.006; T1530 | Retrieved secrets and configuration, validated discovered credentials, and collected accessible S3 objects. |
Vulnerabilities:
- No CVE is associated with this campaign. The actor abused valid credentials and standing cross-account trust.
Malware/Tool:
- TruffleHog was used to validate discovered credentials. Security researchers also identified sso_agent.iso on attacker infrastructure, but did not confirm deployment in victim environments.
Additional Sources
Qubole: Creating a Cross-Account IAM Role for QDS
Sentra: Inside the Qubole Supply Chain Attack
Qubole: Configuring Access Settings Using IAM Roles
↑
Share