← Back to Blog
Hot TopicOn This Site

The Expedia Breach and the Qubole AWS Attack (November 2026): How to Audit Vendor Access to Your Cloud

Expedia confirmed on 12 November 2026 that attackers got in through a compromised third-party data platform provider. Days earlier, Wiz and Sentra documented an attacker using credentials from Qubole's AWS account to walk into customer AWS environments through old, trusted integration roles. Here is what is confirmed, what is only claimed, and the vendor-access audit every startup on AWS should run this week.

·11 min read

On 12 November 2026, Expedia confirmed to Skift that attackers reached its systems through a compromised third-party data platform provider, in a supply chain attack that also hit other companies. Expedia has not named the vendor or said what data was taken; an extortion group calling itself n0n claims hundreds of millions of loyalty, profile and account records, which nobody has verified. Three days earlier, Wiz and Sentra (working with Mandiant) published how a very similar attack worked: between 14 and 26 October, someone used IAM credentials from an AWS account associated with Qubole, a big data platform vendor, to assume the integration roles its customers had created, read their secrets and download S3 data. Most of those roles were dormant. If you run on AWS, the action is simple and urgent: list every role that trusts an outside account, delete the ones you no longer use, and lock the rest to the vendor's network.

I'm Kunal Vohra, a fractional CTO and a security researcher and cybersecurity advisor. This week I wrote about the wfx-loot GitHub Actions attack, where stolen tokens rewrote AWS trust policies. This one is the quieter version of the same problem: nobody had to rewrite anything. The trust was already there, granted years ago to a vendor, and never taken back.

Who should care

  • Anyone who has ever connected a data or analytics vendor to AWS. Data platforms, BI tools, ETL services, cost tools, security scanners and observability vendors often ask you to create a cross-account IAM role. If you set one up and stopped using the product, it may still work.
  • Former and current Qubole customers. Wiz lists the exact account, principal IDs, session names and IP addresses to search for. Run the check in step 1 below.
  • Startups that keep secrets in Lambda environment variables, Secrets Manager or SSM. Those were the first things the attacker read, and the keys found there were then tested against AWS, Microsoft Entra ID and Okta.
  • Expedia Group customers. Expedia Group brands include Expedia, Hotels.com, Vrbo, Orbitz and Travelocity. Expedia has not said which brands or customers are affected, so treat the next few weeks as a time for extra suspicion of travel emails.
  • Founders who sell to enterprises. Your customers' security questionnaires are about to ask how you manage vendor access. It helps to have an answer before the question arrives.

What actually happened

What Expedia confirmed

Skift's report of 12 November quotes Expedia: "We are aware of a supply chain attack, reported by security researchers". Expedia says the attackers came in through a compromised third-party data platform provider, that it is investigating, and that its platforms remain safe. As of 14 November it had not named the provider, confirmed what data was taken, or said how many people are affected. I found no report of customer notifications or forced password resets.

What the extortion group claims

n0n, which SOCRadar describes as a data-theft and extortion group that appeared in mid-September 2026 and does not use ransomware, listed Expedia on its leak site. Per Security Affairs and Forbes, it claims more than 830 million loyalty programme records from 35 programmes, 58 million traveller profiles and more than 333 million account and credential records, plus source code. Security Affairs alone also lists about 2.5 million card spending records linked to hashed emails; other outlets say there is no confirmed evidence of payment card exposure. These are the attacker's numbers, not Expedia's, and rows are not people. One person can appear in many rows.

What Wiz and Sentra documented

Wiz published "Trusted by Default: Qubole Cross-Account Access Abused in the Wild" on 9 November, and Sentra, which found the activity, published its own analysis the same day. The facts that matter:

  • Window: 14 to 26 October 2026. Wiz also saw separate, unlinked activity from 8 to 11 October using customer-owned AWS keys it believes were created for Qubole integrations.
  • Access: the attacker used IAM credentials from AWS account 805246085872, which Wiz says Qubole's platform uses, to call AssumeRole on integration roles in customer accounts. Every call included the correct External ID and succeeded. How the attacker got those credentials has not been published.
  • Dormant roles: in most victim environments, the attacker's calls were the only role assumptions from that vendor account in October. Sentra says the roles had been dormant since the spring.
  • What they did: listed S3 buckets, databases, clusters, instances, roles and KMS keys; read Lambda function configuration (which returns environment variables), Secrets Manager values and SSM parameters; downloaded S3 objects, in some sessions entire buckets; and used TruffleHog to check whether the harvested keys worked against AWS, Entra ID and Okta. Sentra counted roughly 100 secrets read in about 7 minutes. Neither firm saw write or delete actions.
  • Scale: Wiz does not give a number. Sentra says more than a dozen organisations, and elsewhere "dozens" of Qubole customers.
  • Who: in an update on 11 November, Wiz noted Risky Business's report that n0n claimed the intrusion. Wiz says it has not identified ties to other public activity.

A vendor you stopped paying can still hold the keys to your cloud. Offboarding isn't finished until the trust policy is gone.

Is Expedia's breach the Qubole attack?

Probably, but it is not confirmed. Skift describes Qubole as a onetime Expedia vendor and says Expedia did not confirm a Qubole link. Neither the Wiz nor the Sentra post names Expedia. Security Affairs wrote on 14 November that the Expedia attack used a stolen Qubole vendor key, but that link reads as the outlet's own conclusion. Security Affairs also calls Qubole a Google Cloud partner, while Wiz and Sentra describe an AWS attack. For your own stack it doesn't matter which is right: the technique is documented, and it works on any cloud where you've granted a vendor standing access.

Why this works so well

Cross-account roles are the right way to give a vendor access. They are far better than handing over long-lived access keys. But they have three weak spots that this campaign used:

  1. The External ID is not a secret. It protects against one specific "confused deputy" mistake. Wiz points out it is logged in plaintext in CloudTrail and often stored by the vendor and in infrastructure-as-code. If the vendor's account is compromised, the attacker has it.
  2. Trust outlives the contract. Nobody gets an alert when a vendor role goes unused for six months. The role stays, with whatever broad permissions the vendor's onboarding guide asked for.
  3. Read access is enough. No write actions were needed. Reading Lambda configuration, Secrets Manager and SSM turns one vendor role into credentials for everything else.

I wrote in the WorkOS and ARTEX post that the boring gaps are now found at machine speed. This is the same lesson from a different direction. And as the Microsoft 365 outage post argued, every vendor you depend on is part of your system, whether for uptime or for security.

The vendor-access audit to run this week

The commands below are a starting point I wrote for this post; read them before running, and adjust regions, profiles and dates for your accounts.

1. Check whether you were touched by the Qubole campaign

If you ever connected Qubole, search CloudTrail from 1 October 2026 onward in every account that held an integration role. CloudTrail's event history covers 90 days of management events per region:

for r in $(aws ec2 describe-regions --query "Regions[].RegionName" --output text); do
  aws cloudtrail lookup-events --region "$r" \
    --lookup-attributes AttributeKey=EventName,AttributeValue=AssumeRole \
    --start-time 2026-10-01T00:00:00Z \
    --query "Events[].CloudTrailEvent" --output text \
  | grep -E "805246085872|quiam-|AIDA3W7DIFLYOLTK63SFS" && echo "HIT in $r"
done

Wiz's indicators, in short: role session names starting with quiam-; principal AIDA3W7DIFLYOLTK63SFS (seen only in attacker activity); principal AIDAJYRRH5Y3A7AAJM2HW from a non-AWS IP; and the IPs 57.129.54.120 (OVH, the main one), 45.83.220.211, 185.209.199.73 and 89.37.63.175 (Mullvad VPN exits, so not conclusive alone). Also check your Entra ID and Okta sign-in logs for the TruffleHog user agent. If you have an organisation trail in S3, run the same filter in Athena so you aren't limited to 90 days.

One caveat Wiz stresses: S3 object downloads are not in CloudTrail management events. Without S3 data events or server access logs, you can see that a role was assumed but not what it took. If you find a hit, assume everything that role could read is gone.

2. List every role that trusts an outside account

This is the step every AWS startup should run, Qubole or not:

ME=$(aws sts get-caller-identity --query Account --output text)
aws iam list-roles --output json | jq -r --arg me "$ME" '
  .Roles[]
  | . as $r
  | [ .AssumeRolePolicyDocument.Statement[]?.Principal.AWS? // empty
      | if type == "array" then .[] else . end
      | select(test("^(arn:aws:iam::)?[0-9]{12}") and (contains($me) | not)) ]
  | select(length > 0)
  | "\($r.RoleName)\t\(join(","))"'

For each role, check when it was last used:

aws iam get-role --role-name ROLE_NAME --query "Role.RoleLastUsed"

Write the results into a short register: role, vendor, what it can do, last used, owner. If you use AWS Organizations, IAM Access Analyzer's external access findings give you the same view across accounts.

3. Delete what you don't use

Any vendor role unused for 90 days and not tied to a live contract should go. Detach its policies and delete it. If you're nervous, first replace its trust policy with one that allows nothing, wait a week, then delete. Do the same for any long-lived access keys you once created for a vendor; Wiz's separate 8 to 11 October activity used exactly that kind of key.

4. Lock the roles you keep to the vendor

Wiz's recommendation is specific: trust the vendor's exact role rather than its whole account, use a distinct External ID per role, and add a network condition. In Wiz's words: "In this activity, a source IP condition alone would have blocked every role assumption we observed." A trust policy along those lines:

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Principal": { "AWS": "arn:aws:iam::111122223333:role/vendor-integration" },
      "Action": "sts:AssumeRole",
      "Condition": {
        "StringEquals": { "sts:ExternalId": "a-unique-value-for-this-role" },
        "IpAddress": { "aws:SourceIp": ["203.0.113.0/24"] }
      }
    }
  ]
}

Get the vendor's published egress ranges (or VPC endpoint IDs, in which case use aws:SourceVpce) before you apply this, or you'll break the integration. If a vendor can't tell you where its calls come from, that is useful information for your next renewal.

5. Shrink what vendor roles can read

Most vendor roles need a few buckets or tables, not your whole account. Remove lambda:GetFunctionConfiguration, secretsmanager:GetSecretValue, ssm:GetParameter* and broad s3:GetObject unless the product truly needs them. Better still, move secrets out of Lambda environment variables and into Secrets Manager with resource policies that name the roles allowed to read each one.

6. Turn on the logs you'll need

Use an organisation-wide CloudTrail, enable S3 data events on buckets with customer data, and alert when any role is assumed by an outside account from an unexpected network. GuardDuty helps, but even a simple EventBridge rule on AssumeRole from external accounts will tell you when a dormant vendor wakes up.

7. If you were hit: rotate, then revoke sessions

Rotate every secret, SSM parameter, Lambda environment variable and access key the role could read. Then end sessions that were already issued, because, as Wiz notes, "Rotating a credential does not end temporary sessions that were already issued." The IAM console's "Revoke active sessions" button on a role attaches a policy like this one:

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Deny",
      "Action": "*",
      "Resource": "*",
      "Condition": { "DateLessThan": { "aws:TokenIssueTime": "2026-11-15T00:00:00Z" } }
    }
  ]
}

Preserve your logs before you change anything, tell the vendor, and check your notification duties. If you hold data on Indian users, my DPDP Act explainer covers what the law expects after a breach.

8. Make vendor offboarding a checklist item

Add one line to whatever you do when you cancel a SaaS contract: remove its IAM roles, keys, OAuth apps, webhooks and GitHub app installs. On Google Cloud, that means service account grants and workload identity bindings; on Azure, enterprise apps and service principals. It belongs in your startup security checklist next to "remove ex-employees on day one".

If you're an Expedia customer

Expedia has not asked customers to do anything yet. Sensible steps anyway: change your password if you reused it anywhere, turn on any extra sign-in protection the site offers, and be suspicious of emails about loyalty points, refunds or "account verification", because loyalty data makes phishing look real. Go to the site directly rather than clicking links.

What I'd do as a CTO

This week, I'd give one engineer half a day to run steps 2 and 3 across every AWS account, and I'd expect the list to be longer than anyone remembers. Then I'd make the register part of the quarterly access review, the same meeting where we check who still has admin.

I wouldn't stop using vendors, and I wouldn't go back to handing out access keys. Cross-account roles are still the right pattern. The fix is ownership: every trust relationship has a named owner, an expiry, a network condition, and the smallest read scope that works.

If you're scaling and nobody owns this kind of cloud hygiene yet, that's the gap a fractional CTO closes in the first month. If you just want a second pair of eyes on your AWS trust policies, a technical advisor arrangement covers it. Either way, you can book a 30-minute call and we'll go through your vendor list together.

FAQ

Was Expedia hacked in November 2026?

Expedia confirmed on 12 November 2026 that attackers reached its systems through a compromised third-party data platform provider. It has not named the vendor, confirmed what data was taken, or said how many customers are affected, and it says its investigation is ongoing.

Is the Expedia breach linked to Qubole?

It has been reported but not confirmed. Skift says Qubole was a onetime Expedia vendor and that Expedia did not confirm the link. Wiz and Sentra documented the Qubole campaign without naming Expedia; Security Affairs linked the two in its own reporting.

Did n0n really steal a billion Expedia records?

Nobody has verified it. n0n claims more than 830 million loyalty records, 58 million traveller profiles and over 333 million account records. Expedia has not confirmed these figures, and record counts are not the same as the number of people affected.

How do I know if my AWS account was affected by the Qubole attack?

Search CloudTrail from 1 October 2026 for AssumeRole calls from account 805246085872 from non-AWS IP addresses, session names starting with quiam-, or principal AIDA3W7DIFLYOLTK63SFS. Wiz's post lists the full set of indicators.

Is an External ID enough to secure a cross-account role?

No. The External ID prevents one kind of mix-up between a vendor's customers, but it is not a secret and it was presented correctly in every malicious call here. Add a source IP or VPC endpoint condition, trust a specific vendor role, and keep permissions narrow.

Should Expedia customers change their passwords?

Expedia has not asked customers to. Changing a reused password and watching for travel and loyalty phishing are reasonable precautions while the investigation continues.

Sources

Written By

Kunal Vohra

Kunal Vohra

Technical Co-Founder & Fractional CTO

I've co-founded 6+ startups across India, the UAE, and the US, spanning AI, Web3, fintech, and cybersecurity. I write about the technical and strategic decisions that determine whether a startup thrives or stalls.

Comments

Loading comments…

Enjoyed this article?

More writing on AI, Web3, and building startups.