When a trusted vendor becomes the way in: lessons from the Qubole AWS incident
Between 14 and 26 October 2026, an attacker used credentials belonging to a software vendor to walk into dozens of its customers’ AWS accounts. No AWS service was broken. The attacker simply used access that customers had granted years earlier and, in many cases, forgotten about.
What happened
Qubole is a big-data platform that integrates with AWS. To do that, customers create an IAM role in their own account that trusts Qubole’s AWS account. According to Wiz, which led the investigation, an attacker compromised two IAM users in Qubole’s account and used them to assume those roles in customer environments.
Once inside, the pattern was the same each time. The attacker listed the S3 buckets the role could reach, enumerated their contents and downloaded them. They also pulled secrets from AWS Secrets Manager and SSM Parameter Store, read Lambda environment variables, and tested the keys they found. Sentra, which first spotted the activity, reported dozens of victims and said the attacker collected 100 passwords and keys in seven minutes.
AWS was clear that its services operated as designed. The roles did exactly what their trust policies allowed.
Why this matters if you are a small team
Most start-ups connect several vendors to AWS: monitoring, cost tools, CI/CD, data platforms, security tools. Each one usually asks for a cross-account role. Three things then tend to go wrong:
- Roles outlive the vendor. The trial ends or the tool is replaced, but the role stays.
- Permissions are broad. Read access to every S3 bucket, or to Secrets Manager, is common because it is easier than scoping.
- Nobody watches the role. It is used rarely, so unusual use does not stand out to anyone.
In this incident, several of the roles the attacker used were otherwise unused. A role that does nothing for months and then lists every bucket in the account is a strong signal, but only if something is watching.
Three checks to run today
- List every role that trusts another AWS account. Create an IAM Access Analyzer external access analyser with your own account as the zone of trust; its findings list every role that another AWS account can assume. For each role, check when it was last used:
aws iam get-role --role-name <role> --query Role.RoleLastUsed. Delete any role for a vendor you no longer use. - Tighten the roles you keep. Require an External ID in the trust policy, restrict source IP ranges where the vendor publishes them, and scope permissions to the specific buckets and secrets the vendor needs. Very few vendors need
secretsmanager:GetSecretValueon everything. - Look back, then keep watching. If you ever connected Qubole, review CloudTrail from 1 October for
AssumeRolecalls from its account, then rotate any secret those roles could read. For everyone else, make sure Amazon GuardDuty is enabled in every account and region, so unusual use of credentials raises a finding.
Where bearhug helps, and where it does not
bearhug builds on AWS Security Hub and Amazon GuardDuty. When GuardDuty or Security Hub raises a critical or high-severity finding, bearhug emails you a plain-English explanation of what it means and what to do, rather than a JSON blob in a console nobody opens. The multi-account dashboard puts production, staging and development side by side, so a finding in a quiet account is not missed.
bearhug does not replace the review in step 1. No tool can tell you whether a vendor relationship is still wanted; that is a decision for your team. What bearhug does is make sure that, when AWS sees something unusual, a person on your team hears about it in words they can act on.
Start a 14-day free trial through AWS Marketplace or see the demo.
Sources: Wiz Research, 9 November 2026; BleepingComputer, 10 November 2026.