An unidentified attacker used IAM credentials tied to Qubole’s own AWS account to assume trusted roles inside multiple customer AWS environments between October 14 and 26, 2026, harvesting secrets and downloading S3 data, according to Wiz, Sentra and Mandiant. Many of the roles were dormant, which is why the attack worked.
What happened in the Qubole AWS supply-chain attack
On November 9, 2026, Wiz published a CIRT analysis describing an attacker who used IAM credentials from an AWS account associated with Qubole to assume integration roles in customer AWS environments. Wiz worked with Sentra and Google’s Mandiant. Sentra first spotted suspicious activity across several of its own customers and recognised it as a broader supply-chain attack.
Qubole is a cloud data-processing platform, owned by Idera according to The Hacker News. It works inside customer AWS accounts through cross-account IAM roles, so customers do not hand over long-lived access keys. The attacker abused that trust relationship.
Timeline
| Date (2026) | Event | Source |
| October 8–11 | Separate actor tests customer-owned AWS access keys believed to be created for Qubole integrations; a small number of cloud environments compromised. Not linked to the main campaign. | Wiz |
| October 14 | First observed AssumeRole calls from Qubole account 805246085872 via a Mullvad VPN address | Wiz |
| October 15 | Most activity shifts to an OVH-hosted server | Wiz |
| October 18 | Second IAM user from the same account used; sweep across multiple victim environments | Wiz |
| October 26 | Last observed role-assumption activity | Wiz |
| November 9 | Wiz publishes findings and indicators of compromise | Wiz |
| November 10 | Coverage by CSO Online, The Register and others | Press |
Confirmed, reported and unknown
- Confirmed by Wiz: multiple customer environments were entered; secrets and S3 data were accessed; Wiz calls it a broad, opportunistic supply-chain campaign.
- Reported by Sentra (via CSO Online): the campaign affected dozens of Qubole customers. Wiz has not published a victim count and declined to give one to The Register.
- Not disclosed: how the attacker obtained the Qubole-account credentials, victim names, sectors, and what data was ultimately taken.
Root cause: trust that outlived the integration
The attacker did not break into customer accounts directly. It held credentials that AWS treats as legitimate, because customers had already told AWS to trust Qubole’s account. Two weaknesses made that trust exploitable.
Dormant trust policies. Wiz found that most roles the attacker assumed were not otherwise in use. In a majority of observed environments, the attacker’s requests were the only AssumeRole activity from the vendor account in October. The trust policies were still in place, so the requests succeeded. Dataconomy reports that many organisations left Qubole trust policies active after discontinuing the platform.
An External ID that is not a secret. Every malicious request carried the correct External ID and succeeded. AWS documents the External ID as a guard against the confused-deputy problem, not as a credential. It appears in plaintext in CloudTrail and is often stored by vendors and in infrastructure-as-code. Wiz advises against relying on it as the only control.
Broad vendor permissions. Wiz notes that access to Lambda configuration, Secrets Manager, SSM parameters and wide S3 read is often granted to vendor roles for convenience. The Hacker News adds that Qubole’s setup guide includes a sample S3 policy allowing read access to every bucket in the account. That detail is The Hacker News’s observation, not a Wiz finding.
The unanswered question. No source has said how the attacker got credentials for the Qubole-associated IAM users. Neither Qubole nor AWS had commented publicly when The Register and The Hacker News checked. None of the sources reviewed cite a CVE; this looks like an identity and trust-configuration problem rather than a software vulnerability.
Attack methodology: from AssumeRole to bulk S3 download
Wiz describes the intrusion in five stages. All details below come from the Wiz CIRT report unless noted.
- Initial access via role assumption. On October 14, a Mullvad VPN address called sts:AssumeRole against customer integration roles from Qubole account 805246085872, using IAM user AIDA3W7DIFLYOLTK63SFS. Wiz saw that user only in attacker activity. From October 15 most calls came from an OVH-hosted server, and on October 18 the attacker added a second IAM user, AIDAJYRRH5Y3A7AAJM2HW, which also carries legitimate Qubole traffic.
- Discovery. Role sessions were named with a quiam- prefix, such as quiam-enum. The attacker listed buckets, databases, clusters, EC2 instances, security groups, RDS instances and snapshots, IAM roles and KMS keys.
- Credential harvesting. The attacker read Lambda environment variables with GetFunctionConfiguration, secrets with GetSecretValue, and SSM Parameter Store values with GetParameter.
- Credential validation. Using TruffleHog, the attacker tested the victims’ own AWS keys (GetCallerIdentity), Entra ID client secrets and Okta OAuth client credentials. AWS keys from several accounts were checked in bursts lasting seconds. Sentra, via CSO Online, reports more than 100 passwords and keys read within minutes and says it saw no write operations.
- S3 exfiltration. The pattern was ListBuckets, ListObjects, HeadObject, then GetObject. The attacker was opportunistic, reading whatever each role could reach and sometimes downloading entire buckets in one session.
The sso_agent.iso finding
Via Validin, Wiz found a file named sso_agent.iso on the attacker’s OVH server. Wiz believes it is a bootable Linux image meant to be booted remotely on a victim machine. It runs for 90 minutes, collects system and disk information, polls the server on TCP port 23458 for shell scripts to run, then hard-reboots to erase evidence. It loads the ntfs3 driver, which suggests Windows targets. Wiz has not found any deployment and cannot confirm it was used. Treat it as a capability indicator, not evidence of further compromise.
Why S3 theft may be undercounted
CloudTrail management events record the role assumption and bucket listing, but not object reads. Wiz says several affected environments lacked S3 data events or server access logs, so responders could not tell which objects were read.
Who is behind it? No attribution yet
Confirmed: nothing. Wiz has not tied the campaign to any publicly reported activity or named threat group, and CSO Online says the same.
Expert assessment: Wiz says the targeting and data theft look opportunistic and do not match typical state-backed priorities. It adds that it has no direct insight into the actor’s motive or what will be done with the stolen data. That is a judgement, not an attribution.
Unverified: no ransomware group, extortion site or leak claim connected to this campaign appeared in the sources reviewed. Readers should be sceptical of any such claim until a vendor or responder corroborates it.
A second, unlinked cluster
Wiz separately saw activity from October 8 to 11 using customer-owned AWS access keys believed to be created for Qubole integrations. The actor tested all of them within a single hour on October 8, and in a small number of cases went on to compromise the victim’s cloud environment. Wiz says it cannot link this to the role-assumption campaign. Defenders should treat them as two cases to investigate, not one.
Impact assessment
The full impact is unknown. What the sources support:
- Scope: “multiple” customer environments per Wiz; “dozens” of customers per Sentra, as reported by CSO Online. No victim names or sectors have been published.
- Data: S3 contents, in some cases whole buckets, plus secrets from Lambda, Secrets Manager and SSM Parameter Store. Sentra reports thousands of additional credential records inside downloaded files.
- Downstream credentials: because the attacker tested AWS, Entra ID and Okta credentials found in victim accounts, exposure may extend beyond AWS into identity and SaaS tenants. Whether any of those logins succeeded has not been disclosed.
- Visibility gap: victims without S3 data-event logging cannot prove what was or was not read. That has direct consequences for breach-notification decisions.
- Operational disruption: none reported. Sentra says it saw no write operations; this looks like theft, not sabotage.
- Financial, reputational and regulatory: no loss figures, fines, lawsuits or regulator statements have been published, so none are given here.
Customers whose buckets held personal, regulated or contractual data should consult counsel on notification duties, which vary by jurisdiction and by what log evidence exists.
Why industrial and OT security teams should care
What is known: no source names an industrial, utility or manufacturing victim, and no OT network, ICS, SCADA or IoT device was reported compromised. This was an AWS cloud intrusion. The OT relevance below is OT Ecosystem’s analysis, not a finding from Wiz, Sentra or Mandiant.
Why it still matters: industrial organisations increasingly send plant telemetry, historian exports, quality data and maintenance records to cloud data lakes and analytics platforms. Those pipelines depend on the same pattern seen here: a vendor identity that can assume a role in the owner’s cloud account. If the integration is retired but the role remains, the access path remains too.
- Data exposure: S3 buckets holding process data, asset inventories, network diagrams, engineering files or remote-access documentation are valuable reconnaissance material, even without touching a controller.
- Secrets in cloud config: Lambda variables, Secrets Manager and Parameter Store often hold credentials for IT/OT integration services, data gateways and remote-access tools. The attacker harvested exactly these categories.
- Identity pivot: Entra ID and Okta credentials found in cloud accounts can open the corporate identity layer that many OT remote-access and jump-host designs depend on.
- Governance gap: OT asset owners tend to inventory controllers and firewalls carefully, but cloud trust relationships created during a pilot or proof of concept often sit outside that inventory.
Practical test for OT teams: ask your cloud and data-platform teams for a list of every IAM role that trusts an external AWS account, who owns it, and when it was last used. If no one can answer, that is the finding.
Response, recovery and latest developments
Vendor response: as of November 10, neither Qubole nor AWS had responded to The Register, and The Hacker News found no statement from Qubole or Idera on their websites. The researchers also did not say whether the credentials have been disabled or whether the activity has stopped. OT Ecosystem will update this article when a statement appears.
Detection and disclosure: Sentra detected the activity in customer environments, shared it with Wiz, and Wiz brought in Mandiant. Wiz published indicators of compromise, CloudTrail hunting queries and recommendations for both customers and SaaS providers.
Recovery status: unknown. Wiz’s guidance implies that victims must rotate every secret the role could read and revoke active role sessions, because rotating a credential does not end temporary sessions already issued.
Latest points to watch:
- an official Qubole or Idera statement, including how the credentials were obtained and whether they are revoked;
- any AWS guidance for customers with Qubole trust policies;
- a published victim count or sector breakdown;
- evidence that sso_agent.iso was deployed;
- any attribution or extortion claim.
Security recommendations
These steps follow the Wiz CIRT guidance, with OT Ecosystem’s added emphasis on inventory and ownership.
Hunt now (any organisation that ever connected Qubole)
- Search CloudTrail from October 1, 2026 in every account that holds an integration role, including accounts where you believe Qubole is unused.
- Look for AssumeRole events where userIdentity.accountId is 805246085872 and sourceIPAddress is not an AWS address. Wiz saw all legitimate Qubole traffic come from AWS infrastructure.
- Look for AssumeRole by principal AIDA3W7DIFLYOLTK63SFS (any source), and by AIDAJYRRH5Y3A7AAJM2HW when the source is not AWS.
- Search for role sessions named quiam-* and for GetCallerIdentity calls with a TruffleHog user agent. Check Entra ID and Okta sign-in logs for the same user agent.
- For every session found, review later Secrets Manager, SSM, Lambda and S3 activity.
Contain
- Rotate every secret, parameter and Lambda environment variable the role could read, and revoke the role’s active sessions. Rotation alone does not end issued sessions.
- Remove Qubole roles you no longer use, or remove Qubole’s account from their trust policies.
- Rotate the External ID in coordination with Qubole, using a distinct value per role where supported.
- Rotate downstream Entra ID and Okta credentials if they were stored in reachable AWS locations.
Harden
- Replace the account-wide :root principal with the specific IAM principal the vendor uses, where supported.
- Add aws:SourceIp or aws:SourceVpce conditions limited to documented vendor ranges. Wiz says a source-IP condition alone would have blocked every role assumption it observed.
- Treat the External ID as non-secret and never the sole control.
- Apply least privilege to vendor roles; remove blanket S3 read and access to Secrets Manager, SSM and Lambda configuration unless the integration needs it.
- Alert on vendor-account AssumeRole from a VPN or hosting provider the vendor has not used before.
- Use an organisation-wide CloudTrail covering all accounts and regions, with S3 data events or server access logs on sensitive buckets.
- Keep an owned inventory of external-account trust relationships, and review it on contract end, pilot end and staff turnover.
If you are a SaaS provider
Wiz advises assuming customer roles from workload roles with short-lived credentials rather than long-lived IAM user keys, restricting where those principals can be used, publishing egress ranges so customers can pin trust policies, and monitoring your own CloudTrail for AssumeRole calls into customer accounts from unexpected networks.
Indicators of compromise (from Wiz)
| Type | Indicator | Note |
| AWS account | 805246085872 | Source of observed role assumptions; also used for legitimate Qubole activity |
| IAM principal ID | AIDA3W7DIFLYOLTK63SFS | Seen only in attacker activity |
| IAM principal ID | AIDAJYRRH5Y3A7AAJM2HW | From Oct 18; also carries legitimate traffic, so filter by source |
| IP | 57.129.54[.]120 | OVH; primary address |
| IP | 2001:41d0:70d[:]7800:: | OVH; credential checks |
| IP | 45.83.220[.]211 | Mullvad VPN; shared exit node |
| IP | 185.209.199[.]73 | Mullvad VPN; shared exit node |
| IP | 89.37.63[.]175 | Datacamp, registered to Mullvad VPN; shared exit node |
| Session name | quiam-* | Role session prefix |
| File SHA-256 | 65f061281c9a712e9502793dc4c5035cae490b84da05f7d4223cefa65ad396e4 | sso_agent.iso |
VPN addresses are shared, so a match alone is not conclusive. Wiz also lists several Boto3 and AWS CLI user agents, which it says are suspicious only alongside another indicator; see its report for the full list.