FACT: Wiz and secondary reporting support this minimum path: IAM credentials associated with Qubole’s AWS account 805246085872, including principal ID AIDA3W7DIFLYOLTK63SFS, called sts:AssumeRole against customer roles that still trusted Qubole. Where those roles required an External ID, successful assumption means the request satisfied that condition; External ID was not a second authentication factor. Some former customers had left that trust active. UNKNOWN: The evidence does not establish compromise of Qubole’s entire AWS account, how the credentials and External IDs were obtained, or that every customer integration was accessed.
Tonight, query organization-wide CloudTrail from at least October 14–26—and farther back where retained—for eventSource=sts.amazonaws.com, eventName=AssumeRole, and either userIdentity.accountId=805246085872 or userIdentity.principalId beginning with the cited ID. For every match, preserve eventTime, recipientAccountId, requestParameters.roleArn, requestParameters.externalId, sourceIPAddress, userAgent, awsRegion, principalId, session name, and the resulting assumed-role ARN. Treat non-AWS or otherwise unexpected source IPs as especially suspicious. Pivot on that role session into lambda.amazonaws.com configuration reads, Secrets Manager GetSecretValue, SSM parameter reads, and S3 listing or GetObject activity; also hunt for persistence such as UpdateAssumeRolePolicy, PutRolePolicy, AttachRolePolicy, CreateAccessKey, or new principals. S3 object reads may be invisible if data events were not enabled, so an empty query is not necessarily clean.
In IAM, inventory every trust policy naming account 805246085872 or any Qubole principal, including roles in former accounts and IaC templates that could recreate them. Record the role’s attached and inline permissions, permission boundary, External ID condition, and reachable secrets or buckets. Former customers should remove the trust immediately. Current customers should suspend it until Qubole provides a newly established principal and External ID through a verified channel, then rebuild it with least privilege and source restrictions where operationally possible. AWS STS appears to have enforced the policy presented to it; vendor credential protection is Qubole’s responsibility, while the trusted principal, role permissions, logging, and abandoned integration are customer controls. “Shared responsibility” remains a remarkably efficient way to say two parties now have incident tickets.
Escalate from trust removal to full incident response on any successful unexpected AssumeRole, even without confirmed downstream reads, because temporary credentials were issued. Also escalate for secret, parameter, Lambda-environment, or S3 access; policy or principal changes; suspicious follow-on use of exposed AWS, Entra ID, or Okta credentials; or logging gaps that prevent scoping a successful session. Preserve evidence first, revoke role sessions, rotate every credential reachable by that role, and investigate downstream identities. NotC2 does not materially change that threshold: it is separate proof-of-concept detection context showing AWS services can carry C2-like traffic, not evidence linking that framework to the Qubole activity.