al Stored Secrets
Cybersecurity researchers have uncovered a sophisticated and targeted cloud intrusion campaign in which malicious actors compromised Amazon Web Services (AWS) accounts belonging to multiple customers of the big data platform Qubole. According to findings published today by cloud security firm Wiz, the unauthorized access occurred between October 14 and October 26, allowing the attacker to infiltrate victim environments, harvest sensitive stored credentials, and siphon valuable data from cloud storage buckets.
The breach centered on compromised credentials associated with an AWS account operated by Qubole, a prominent data lake platform owned by Idera. Qubole’s architecture typically requires deep integration within its customers’ AWS environments to function smoothly. To establish this operational bridge, customers configure a dedicated AWS Identity and Access Management (IAM) role—a specialized identity with a defined set of permissions tailored to allow the platform to execute tasks on their behalf.
The security mechanism governing this relationship relies on a trust policy embedded within the IAM role, explicitly permitting Qubole’s primary AWS account to assume the role and inherit its associated permissions. Crucially, every time this cross-account access is invoked, the request—known technically as an AssumeRole operation—must present a matching identifier code provided by Qubole, commonly referred to as an External ID.
Investigators from Wiz discovered that the attacker successfully leveraged credentials tied to Qubole’s overarching platform account to execute these cross-account requests. While the exact vector and method by which the threat actor managed to acquire these foundational credentials remain publicly undisclosed, the mechanics of the attacks revealed a systematic exploitation of trust relationships. Intriguingly, Wiz noted that many of the specific IAM roles targeted and assumed during the campaign were otherwise dormant or unused within the victims’ environments. Nevertheless, because these roles still maintained active trust relationships pointing to Qubole’s platform account, the unauthorized AssumeRole requests were accepted by AWS infrastructure.
The initial indicators of this activity were flagged by cloud security firm Sentra, which spotted anomalous behaviors across several customer environments before alerting Wiz researchers for deeper technical analysis. Following this initial notification, security experts began mapping out the full scope of the multi-week campaign.
Why the Roles Let the Attacker In
A technical analysis of the incident reveals how the attacker bypassed administrative suspicion during the operation. Every single AssumeRole request executed by the threat actor carried a valid External ID and originated from the trusted account identifier, satisfying the baseline authorization requirements established by AWS. Under standard cloud architecture guidelines, an External ID is designed to prevent the "confused deputy" problem in third-party service integrations; however, it is not treated as a cryptographic secret. According to official AWS documentation, anyone who possesses sufficient administrative access to the role within the third-party platform or the trusting account can view the External ID, making it vulnerable if the controlling platform’s security perimeter is breached.

Once inside the compromised environments, the attacker did not immediately focus on specific high-value targets. Instead, the primary objective appeared to be broad reconnaissance and the mass extraction of stored authentication material. Investigators observed the threat actor systematically scouring victim environments for hard-coded or stored secrets. This included reading environment variables embedded within Lambda functions, pulling sensitive credentials directly from AWS Secrets Manager, and extracting configuration values from the SSM Parameter Store.
To maximize the utility of the harvested data, the attacker automated the credential-testing process. The stolen material—ranging from proprietary AWS access keys and Microsoft Entra ID client secrets to Okta administrative credentials—was rapidly processed through TruffleHog, an open-source secret-scanning tool, to verify whether the keys were active and valid for further exploitation.
Beyond credential harvesting, the campaign involved significant data exfiltration from Amazon Simple Storage Service (S3). Rather than deploying targeted searches for specific proprietary files or databases, the intruder engaged in indiscriminate retrieval. The attacker downloaded data from virtually any S3 bucket accessible to the compromised IAM role, in some instances copying entire storage buckets in a single continuous session.
Security researchers have emphasized that this campaign does not align with traditional tactics, techniques, and procedures typically associated with state-backed advanced persistent threat (APT) groups. Furthermore, the activity has not been formally linked to any other previously disclosed public cyberattacks or campaigns.
Adding a layer of complexity to the investigation, Wiz researchers also identified a separate, parallel wave of suspicious activity targeting Qubole customers earlier in October. Between October 8 and October 11, a separate attacker leveraged customer-owned AWS access keys that had been explicitly provisioned for Qubole integrations. This secondary vector resulted in the direct compromise of several cloud environments. However, security analysts have found no evidence connecting this earlier access-key abuse to the later, more widespread AssumeRole cross-account campaign.
How to Check Your Accounts
Organizations that have ever integrated Qubole into their AWS infrastructure—even if they have since decommissioned the service or terminated their relationship with the vendor—are strongly urged to conduct comprehensive forensic reviews. Investigators advise security teams to inspect their AWS CloudTrail management logs dating back to October 1.
The primary indicator of compromise to search for is any AssumeRole request originating from the specific AWS account associated with Qubole’s platform, designated by the identifier 805246085872, where the request originated from an IP address or network space outside of the expected AWS infrastructure boundaries.

For every suspicious session identified through CloudTrail logs, security personnel must perform a thorough audit of all subsequent actions taken by that session. Particular attention should be paid to activities involving AWS Secrets Manager, SSM Parameter Store, AWS Lambda configurations, and S3 data access.
A notable challenge faced by incident responders during the investigation was a lack of adequate logging configurations in several victim environments. While standard CloudTrail management events successfully log every AssumeRole invocation and broad S3 bucket-listing operation, they do not inherently record which individual objects within an S3 bucket were read or downloaded. Determining the precise scope of data exfiltration requires detailed S3 data events or comprehensive server access logs. Because several affected organizations lacked these granular logging mechanisms, incident responders were ultimately unable to verify the full extent of the data viewed or stolen by the attacker during certain sessions.
How to Close the Route
To mitigate ongoing risks and close the potential attack vectors exploited during this campaign, cloud security experts have outlined essential remediation steps for affected organizations.
First, businesses must immediately review all third-party IAM roles and cross-account trust policies within their AWS environments. Any integrations that are no longer actively required—including legacy connections linked to platforms like Qubole—should be disabled or permanently deleted to shrink the organization’s overall attack surface.
Organizations that must maintain active integrations with third-party data platforms should audit their IAM trust policies to ensure they adhere to the principle of least privilege. This includes restricting trust relationships wherever possible and strictly monitoring the usage of External IDs and associated IAM roles.
Furthermore, security teams should rotate all sensitive credentials, API keys, and access secrets that may have been exposed during the active window of the breach. This includes updating environment variables in Lambda functions, revoking compromised AWS access keys, and cycling secrets managed within AWS Secrets Manager and SSM Parameter Store. Finally, enterprises are advised to ensure that granular logging features, such as S3 data events and comprehensive CloudTrail tracking, are fully enabled across all cloud environments to ensure future visibility and rapid incident response capabilities.
