The short version
- The breach. An attacker held credentials for the AWS account Qubole uses to reach customer environments, and used them to assume IAM roles that customers had configured to trust that account. Wiz began investigating after Sentra reported suspicious activity in its customers' environments; Mandiant took part.
- What was taken. In the accounts examined, the attacker read Lambda environment variables, Secrets Manager secrets and SSM Parameter Store values, then ran the TruffleHog tool against the AWS keys, Microsoft Entra ID client secrets and Okta credentials it found. It also downloaded data from S3 buckets the roles could reach.
- Who is affected. BleepingComputer reports tens of companies affected. Sentra says dozens of victims and millions of files taken from one customer alone; Wiz has not published a victim count.
- No vendor statement. Neither Qubole nor its owner Idera had posted a statement about the attack when both outlets checked. BleepingComputer said it had emailed Qubole and was awaiting a reply.
- A second, unlinked activity. Between 8 and 11 October an attacker used customers' own AWS access keys, believed to have been set up for Qubole, and compromised a few victims' cloud environments. Wiz has not linked that activity to the role abuse.
Timeline
| Date | Event |
|---|---|
| 8-11 Oct | An attacker uses customers' own AWS access keys, believed to have been configured for Qubole, and compromises a few victims' cloud environments; Wiz has not linked this to the role abuse. |
| 14-26 Oct | An attacker assumes IAM roles that trust Qubole's AWS account, takes stored secrets and downloads data from S3 in multiple customer environments. |
| 9 Nov | Wiz publishes its findings; neither Qubole nor Idera has posted a statement. |
| 10 Nov | BleepingComputer reports the campaign, says tens of companies were affected and that it has asked Qubole for comment. |
How the access worked
Qubole, a data lake platform owned by Idera, runs inside customers' AWS accounts. A customer creates an IAM role whose trust policy names Qubole's AWS account, 805246085872, and the platform assumes that role using AWS's AssumeRole call, presenting a matching External ID. The attacker had credentials for Qubole's account and the correct External IDs: the code is logged in plain text and is not treated as a secret.
BleepingComputer, citing Wiz, reports that two Qubole IAM users were compromised. Most of the roles the attacker assumed were unused, but they still trusted Qubole's account, so the requests succeeded. Wiz says a network restriction on where AssumeRole requests may come from would have blocked every role assumption it observed: "a source IP condition alone would have blocked every role assumption we observed."
What the attacker did inside the accounts
The pattern was the same across environments, according to Wiz: assume a role, list the buckets it could reach, enumerate and inspect objects, then download them. The attacker took whatever was accessible rather than hunting for particular data, and sometimes pulled an entire bucket in one session. Sentra says it collected 100 passwords and keys in seven minutes.
Discovery did not stop at AWS. The stolen AWS keys, Entra ID client secrets and Okta client credentials were tested with TruffleHog to see whether they worked. Several victims lacked S3 data events or server access logs, so responders could not establish which objects had been read.
A separate access-key campaign
Wiz describes a second activity from 8 to 11 October in which an attacker used customers' own AWS access keys, which the researchers believe had been set up for Qubole, and compromised a few victims' cloud environments. The sources do not link this to the AssumeRole abuse and treat the two as distinct.
Wiz has not connected the role-abuse campaign to any other publicly reported activity, and says the targeting does not resemble what state-backed groups usually pursue. Neither source names a culprit or states a motive.
What is not known yet
- Who carried out the intrusion, and whether the goal was extortion, resale of access or something else.
- How the credentials for Qubole's AWS account and the two Qubole IAM users were obtained.
- The exact number of affected customers, and which objects were read from S3 where logging was missing.
- Whether the compromised credentials have been revoked and whether the activity has stopped.
- Whether Qubole or Idera has notified affected customers or confirmed the incident.
What to do
If your AWS account trusts Qubole's account
- Search CloudTrail from 1 October for AssumeRole requests from account 805246085872 where the source address is not an AWS address, and for IAM principal IDs AIDA3W7DIFLYOLTK63SFS and AIDAJYRRH5Y3A7AAJM2HW, session names beginning quiam-, and the OVH and Mullvad addresses Wiz lists — a VPN match alone is not proof.
- For every suspicious session, review what followed in Secrets Manager, SSM Parameter Store, Lambda and S3.
- If a role was assumed, rotate all secrets, parameters and Lambda environment variables it could reach and revoke the role's active sessions, since rotation does not end sessions already issued.
If you keep third-party roles in your cloud
- Remove Qubole roles you no longer need, or delete Qubole's account from the trust policy.
- Replace an account-wide root principal with the specific IAM principal the vendor uses, where the vendor supports it.
- Add an aws:SourceIp or aws:SourceVpce condition limiting AssumeRole to the vendor's documented network ranges.
- Give each role a different External ID and narrow its permissions, including sample policies that allow reading every bucket in the account.
For everyone else
- Enable an organisation trail covering every account and region, plus S3 data events or server access logs for buckets holding sensitive data.
- Alert on AssumeRole requests from vendor accounts that arrive through a VPN or an unfamiliar hosting provider, and check identity provider sign-in logs for the TruffleHog user agent.
Update log
- First published.
Sources
New incident reports by email
One email a week with the reports published since the last one.