An AWS role silent since spring came back to life on October 14. The request came from Qubole's trusted AWS account, carried the correct external identifier, and used a session name beginning with quiam-. To AWS, it looked like the vendor access the customer had authorized. About a second later, temporary credentials opened the environment. No software flaw was required. A permission left behind months earlier still worked exactly as designed.
Over the next 12 days, two IAM users from the same vendor account assumed roles across multiple Qubole customer environments. They mapped storage, compute, databases, and identity settings, then read the secrets stored there. Some victims had stopped using Qubole months earlier. The access had survived the vendor relationship, and the data behind it had kept growing.
On November 12, Expedia Group told Skift threat actors had reached its systems through a compromised third-party data platform provider in a supply chain attack affecting multiple companies. Three days earlier, an extortion threat actor calling itself n0n had named Expedia on its dedicated leak site (DLS), claiming nearly a billion traveler records. Expedia has neither named the provider nor said what data, if any, was taken.
A Key Nobody Remembered Handing Out
Qubole is a cloud data lake platform owned by Idera. To work inside a customer's AWS environment, it asks the customer to create an IAM role trusting Qubole's own AWS account, guarded by an external identifier. When a customer stops using Qubole, the role stays in place unless someone deletes it. SecurityWeek reports more than a dozen organizations were hit, and the abandoned roles could read most of the data the threat actor targeted. I found no public statement from Qubole or Idera.
Wiz traced the activity to IAM credentials for AWS account 805246085872, which Qubole's platform uses. The threat actor began from a Mullvad VPN address on October 14 and worked mostly from an OVH server from October 15 through October 26, adding a second IAM user from the same account on October 18. Every attributed role assumption succeeded. Wiz says many victim environments saw no legitimate vendor traffic during the activity.
Sentra recorded about 100 passwords and keys read in seven minutes from AWS Secrets Manager, Systems Manager Parameter Store, and Lambda environment variables. The threat actor ran TruffleHog to test the stolen secrets. It validated AWS keys, Microsoft Entra ID client secrets, and Okta client credentials. It then pulled data from S3, in some cases whole buckets.
Wiz also found a bootable disk image, sso_agent.iso, hosted on the threat actor's server. It collects system information, polls for shell scripts, and reboots to erase its traces. Wiz has seen no deployment. The image still matters for what it implies about intent, since running code on victim machines is an ambition beyond copying buckets.
Between October 8 and 11, a separate threat actor tested customer-owned AWS keys apparently created for Qubole integrations. It tested all of them within one hour on October 8 and compromised some environments. Wiz cannot connect the two activities. Two parties holding different forms of Qubole-related access in the same month raises the possibility of a common upstream exposure, but the evidence does not locate it. The keys could have escaped from Qubole, from customer-side integration records, or through an unrelated credential market.
What n0n Says It Took
n0n first listed Expedia anonymously as Company #122 on November 3, then named it on November 9 and marked the listing "not paid, exposed." The threat actor claims about 830 million loyalty program rows across 800 files, a 58-million-row traveler profile table, and more than 333 million account and credential records. It dates one database backup, a 174 GB workforce-planning file, to October 21, inside the window Wiz documented.
n0n also claims Vrbo reservation and payment records, roughly half a million guest records, and card-spend and car-rental data. It lists about 747 GB from five more cloud accounts, plus credentials for Expedia's big data platform, source code repositories, and a production cloud environment. According to Security Affairs, the loyalty data covers 35 programs, with email addresses, member IDs, and points balances.
The claimed credentials could prove more consequential than the rows. If valid and sufficiently privileged, those credentials could enable re-entry or code tampering long after any publication. Expedia told Skift it contained the issue once it became aware of it. It has not said whether it rotated those credentials or found evidence of their use. n0n claims one of them is a source code access token reaching 19,935 private repositories.
When n0n named Expedia, CyPro found no sample records or verified data, and a DLS listing is a claim, not a breach. Ransom negotiator Alex Waintraub told Forbes the listing "looks like a pressure tactic by the threat actor."
The picture changed within days. Cyber Daily reported on November 12 the dataset had been published on the dark web as an extract of the full claim. The archive holds 11 folders listing hundreds of thousands of email addresses tied to log-in details, hotel stays, communications, and loyalty IDs. Many of the addresses had never appeared in earlier breaches tracked by Have I Been Pwned.
Expedia says its platforms remain safe for travelers and partners and its investigation continues. As of November 16, I found no Form 8-K under Item 1.05, the cyber incident disclosure item. That absence supports no judgment about materiality. The SEC's four-business-day deadline begins when a company determines an incident is material, and four business days had not elapsed even from Expedia's public statement.
Qubole remains the most likely provider, but nobody has confirmed it. Skift reports outside researchers have not identified Expedia among the affected customers, and Expedia has not confirmed the link. The circumstantial case is strong. Expedia moved its data science work to the cloud on the Qubole platform, and Skift calls Qubole a onetime Expedia vendor.
On November 13, Risky Business reported n0n claiming almost a petabyte of data from all Qubole customers after a months-long intrusion. According to the same report, n0n first approached Qubole privately, and Qubole declined to negotiate. n0n listed Expedia anonymously on November 3 and named it the day Wiz's research went public. The sequence fits an extortion chain moving from a vendor refusing to pay to that vendor's customers.
SOCRadar's breach tracker notes Expedia-linked credentials in infostealer logs before the claim and calls the attack chain unconfirmed. Stolen logins do not match Expedia's own description of a compromised data platform provider, so Qubole remains the strongest candidate, with infostealer access a weaker alternative path.
Separating the Claim From the Breach
Rows in loyalty tables repeat members across programs, transactions, and snapshots, so nearly a billion records describes far fewer people. The 333 million credential records hold "password-like tokens," not confirmed passwords. Expedia's One Key program merged the loyalty schemes of Expedia, Hotels.com, and Vrbo, so a single traveler can appear in several historical tables at once.
CloudTrail management events record the role assumption and the bucket listing, but not the object reads behind S3 exfiltration. Wiz says missing logs in several affected environments left responders unable to see which objects the threat actor read. Without S3 data events or server access logs, investigators may be unable to prove n0n's claimed volume or rule it out, leaving the count unsettled in both directions.
Expedia sells travel booking to banks, airlines, and loyalty providers through its partner business, including Delta Air Lines and loyalty platforms powering thousands of programs. If some of the 35 loyalty programs in n0n's claim belong to partners, their members may sit in the stolen tables without ever knowing Expedia handled their bookings, exposed by a company they never chose. Expedia has not said, and n0n has not named the programs.
Loyalty data carries its own eCrime market. Member IDs, email addresses, and points balances are the raw material for account takeover, where criminals drain points into bookings or gift cards. A list of members sorted by balance is also a phishing target list, and fake One Key or Expedia notices about this breach are the obvious lure.
The abandoned roles read the customers' own storage, and those buckets kept filling long after Qubole stopped touching them. n0n dates the Expedia backup to October 21, and its LiveRamp claim describes feeds from October 2026. Cyber Daily found date ranges in the published extract suggesting older records, so the trove likely mixes historical data with recent backups. Both kinds serve fraudsters and anyone mapping where people travel, with whom, and how often.
n0n has previously turned one apparent upstream compromise into dozens of separate victim listings. In October it listed 38 port, shipping, and logistics organizations within 23 minutes. Cydome tied at least 21 of them to customers of one South Korean maritime software vendor, CyberLogitec, with sample files likely drawn from a single support and project archive. If Cydome is right, one vendor compromise became 38 listings. The Qubole claim follows the same pattern at a larger scale.
n0n named identity resolution company LiveRamp on November 11, claiming 2.2 TB of identity graph feeds and 19.2 million hashed email identities. It claimed DraftKings advertising audience files from 2020 and VF Corporation shopper chat transcripts. I found no confirmation from any of them, and none has been publicly tied to Qubole. The data types all look like what a data lake holds.
Thin Evidence Behind the Russian Label
Forbes headlined the breach as the work of "Russian cyberattackers," and Security Affairs called n0n a newly formed Russian group. The public evidence for that label is thin. SOCRadar lists Russia on its n0n profile without stating a basis. Its own research found n0n's custom tooling written in English, with the only Cyrillic artifact a localized folder path on one victim's computer.
CYFIRMA lists n0n's origin as unknown with low confidence and notes an Uzbek victim argues weakly against a Russian-speaking profile. The DLS and its listings are in English. None of this rules out Russian threat actors, but a nationality label in a headline carries geopolitical weight, and here it rests on an unexplained profile entry. Attribution is a confidence judgment, and no published evidence supports this one.
What the evidence does show is an AI-assisted extortion threat actor operating at volume, the same shift I examined in a China-nexus campaign run by AI agents. SOCRadar found an exposed n0n server running opencode, an open-source AI coding agent. Its configuration held 154 model entries across seven providers, including models with their safety refusals removed, and gateway spending topped US$7,216. Initial access in those cases relied on credential attacks, including credential stuffing against 305 FortiGate SSL-VPN devices, and SOCRadar found no exploits.
SOCRadar reports demands from 0.5 to 33 BTC per victim and up to US$28 million per organization in the October port wave. n0n requires a 10% upfront payment to open a negotiation room and contacts victims' customers and partners to add pressure. As of its November 15 update, Ransomware.live had recorded 222 n0n listings since September, including 100 added during November and 100 naming US organizations. The pricing and pressure are standard eCrime, and Wiz describes the Qubole targeting as opportunistic and unlike state-backed priorities.
Travel Data Has Always Been an Intelligence Target
Whatever n0n wants, the data it claims has a long record as an espionage target. In 2018, people familiar with Marriott's investigation told Reuters Chinese hackers were suspected in the Marriott Starwood breach. The New York Times reported a suspected link to China's Ministry of State Security, and Secretary of State Mike Pompeo appeared to confirm China's role on television. The stolen guest records had not appeared for sale at the time. The Times noted Marriott's role housing US government and military personnel.
In 2020, the US Treasury sanctioned Rana Intelligence Computing, which it said worked for Iran's Ministry of Intelligence and Security. Industry tracks the activity as REMIX KITTEN (aka APT39). Treasury said Rana targeted or compromised about 15 US companies, primarily in the travel sector, giving the ministry a way to track individuals it considered a threat. Norway's security authority now says data from the Amadeus breach has already leaked.
A traveler profile combines a home airport, loyalty IDs, booking history, payment habits, and co-travelers. Together they reveal pattern of life, showing where a person lives, who they move with, and when they leave routine. I showed in October how metadata alone burned the covers of 39 ICE agents with no field ever saying undercover. Travel history is richer metadata. The same logic ran through the Chainalysis breach earlier this month, which took identity data on the investigators who track crypto crime.
In 2020, Bellingcat identified the FSB officers who trailed Alexei Navalny partly through passenger manifests and phone records, drawing on a Russian black market where Telegram bots sold personal data for about 10 euros. Commercial data stolen or sold in bulk becomes counter-intelligence material for whoever holds it, and every intelligence service knows its own officers fly commercial too.
Foreign services already collect visa, border, airline, and telecom records. Expedia data could add booking intent, travel companions, loyalty identifiers, cancellations, and trips never taken. Border records capture only completed journeys. Joined with existing holdings, booking records could confirm identities and reveal plans before a border crossing creates a government record.
Competing Hypotheses
Four explanations fit the public evidence. Each draws on the same facts and diverges on intent. They are not mutually exclusive, and the fourth does not depend on the others.
Hypothesis 1, Opportunistic Extortion
n0n swept every Qubole customer it could reach, copied what it found, and now monetizes each tenant as a separate victim. The volume listings, the CyberLogitec precedent, the ransom pricing, Wiz's assessment, and n0n's own claim of taking data from all Qubole customers support this. Nothing in the role assumptions shows selection. The threat actor took what the roles could read.
Hypothesis 2, Collection Behind an Extortion Front
A state service could run or sponsor an extortion brand to collect bulk data while blaming criminals. The targets fit intelligence interests unusually well. Expedia, LiveRamp, and DraftKings together hold travel, identity resolution, and advertising data. Those are the building blocks of an identity graph. The bootable image suggests interest in deeper access.
The portfolio may also be an artifact of Qubole's customer base, since platforms built to centralize analytics naturally concentrate travel, advertising, and identity records. Against this hypothesis, n0n is loud, public, and indiscriminate. State services rarely advertise their collection on a DLS with a countdown.
Hypothesis 3, Inflated Claims
n0n could be overstating what it holds to raise pressure, counting rows instead of people and listing data it never fully exfiltrated. The published extract is a fraction of the claim, so the headline figures still rest on n0n's word, and n0n profits from fear whether or not they are true. Expedia's confirmation of a real breach and the extract's contents weaken this hypothesis but do not settle the figures.
Hypothesis 4, Downstream Capture
A threat actor's motive does not cap what happens to the data. n0n has already published an extract, and any foreign service could acquire it. If n0n sells the rest privately, buyers choose themselves. Bellingcat's Navalny investigation shows how freely bulk data moves between criminals, investigators, and intelligence officers. Under this hypothesis, eCrime theft becomes intelligence input whether or not anyone planned it.
Where the Data Goes Next
Washington has spent more than two years trying to keep this kind of data out of adversary hands. The Justice Department's bulk data rule took effect on April 8, 2025. It restricts data brokerage and other transactions giving six countries of concern access to bulk sensitive data on Americans. China, Russia, Iran, and North Korea are on the list. The rule governs companies selling and sharing data. Theft walks around it.
The intelligence community already knows what this data reveals because it buys similar data itself. A 2022 report, declassified in 2023, found commercially available information "can reveal sensitive and intimate information about individuals." Adversary services reach the same conclusion. Where Western agencies buy, adversaries can wait for a breach and download.
One leak of travel data identifies trips. Joined with an identity graph linking hashed emails to named people, device IDs, and households, the same travel data turns trips into identities. n0n claims both within the same month, and LiveRamp's claimed files describe hashed emails already resolved to names. Whoever assembles those tables gains more than either breach alone.
For organizations with people in sensitive roles, the practical step is to treat Expedia-linked travel history as potentially exposed until the investigation establishes the scope. Security teams briefing executives, diplomats, and cleared staff should plan for the possibility a foreign service could eventually acquire past trips, frequent routes, and travel companions booked through Expedia brands. The OPM breach showed how long personnel data stays useful to an adversary after the headlines fade.
Offboarding Is a Security Control
One failure here is a permission nobody revoked. Customers stopped using Qubole, and the trust they granted kept working. I examined the same flaw in Denmark, where a company's trusted access pulled 8.8 million identities without malware, and in the ConnectWise compromise, where a vendor's identity opened customer tenants.
Customer offboarding explains the dormant roles. It does not explain how IAM credentials from Qubole's account escaped or why they kept working from VPN and hosting networks. Wiz's advice to vendors covers that side, with short-lived workload credentials in place of long-lived keys and published egress ranges customers can pin. Some affected roles may also have served integrations still in use, where offboarding never applied.
For customers, Wiz's guidance starts with an inventory of every role trusting an external account. Delete trust for integrations no longer in use, and restrict vendor principals by source IP or VPC endpoint. Wiz notes a source IP condition alone would have blocked every role assumption it observed. External identifiers appear in plaintext in CloudTrail and work as labels, not secrets.
Sessions issued before a credential rotation keep working until they expire, and secrets read from Secrets Manager or Lambda variables stay valid until rotated at the source. Defenders holding Qubole roles need to review CloudTrail from October 1 for role assumptions from Qubole's account outside AWS ranges, then rotate every secret those roles could read.
The same week Cursor disclosed how a stolen Fireworks AI vendor key led to backdoored extensions on its marketplace, Expedia confirmed a breach through a third-party provider, most likely one it no longer used. Neither attack needed an exploit. Both used trust a customer had granted and forgotten. A vendor relationship ends when the contract ends, and the access ends only when someone deletes it.
Expedia's breach joins two problems usually handled by different teams. Security teams see a forgotten vendor role and a rotation checklist. Intelligence services see a claimed travel and identity dataset already partly published and available in full to anyone willing to pay n0n. The extortion will end with a payment, a leak, or silence. Any authentic data could keep circulating long after, useful to any service needing to know where people go.
I assess with high confidence the Qubole campaign stole data from multiple Qubole customers, including former customers whose cross-account roles remained active. I assess with moderate confidence Qubole was the provider in Expedia's disclosed incident.
I also assess with moderate confidence n0n holds at least some Expedia data, while its row counts overstate the number of people affected. I assess with low confidence a state directs or sponsors n0n, and with moderate confidence any authentic Expedia data n0n holds will reach at least one foreign intelligence service through publication or sale.
Somewhere in a former Qubole customer's AWS account, a role created for a platform nobody uses anymore is likely still waiting for its next request. At Expedia, the answer to that request may have been a dataset Expedia has not verified, held by a threat actor nobody can place, built from the travel histories of people who never heard of Qubole. Whoever reads those records next, whether a fraudster or a foreign service, will owe that access to a deletion nobody made.
Analyst Annex
Evidence status as of November 16, 2026. The records cover two activity sets kept separate here. The first is the Qubole cross-account role campaign of October 14 to 26, documented by Wiz and Sentra. The second is n0n’s extortion claim against Expedia Group, listed anonymously on November 3 and named on November 9. n0n has claimed responsibility for the Qubole intrusion, but no independent source has attributed the campaign to n0n.
Expedia has confirmed a breach through a third-party data platform provider but has not named it. No researcher has publicly placed Expedia among the Qubole victims. The connection between Expedia, Qubole, and n0n therefore appears under Intelligence Gaps rather than as an established intrusion chain. The Indicators section contains a capped set from the Qubole campaign. The n0n infrastructure documented by SOCRadar predates the Expedia listing and appears only as related context.
Reported marks activity a researcher or news outlet documents directly. Claimed marks n0n’s own statements, none of which has been independently verified. Confirmed marks facts the affected company itself has acknowledged.
Indicators
This capped set covers the Qubole cross-account role campaign. Mullvad addresses are shared VPN exits and should support, not drive, a hunt. Boto3, AWS CLI, and TruffleHog user agents are also suspicious only alongside the listed network or identity indicators. Wiz provides the complete indicator set and hunting guidance.
| Indicator | Context | Status |
|---|---|---|
| IP Addresses | ||
57.129.54[.]120 |
Primary OVH address for role assumption, enumeration, and credential harvesting from October 15 to 26. Also hosted sso_agent.iso
Source: Wiz
|
Reported |
2001:41d0:70d[:]7800:: |
OVH IPv6 address used for credential enumeration and validation from October 16 to 24 Source: Wiz | Reported |
45.83.220[.]211 |
Mullvad VPN exit used to assume customer roles on October 14 and 15. Shared VPN infrastructure, not conclusive alone Source: Wiz | Reported |
185.209.199[.]73 |
Mullvad VPN exit used to assume roles across multiple victim environments from October 18 to 21. Shared VPN infrastructure, not conclusive alone Source: Wiz | Reported |
89.37.63[.]175 |
Datacamp address registered to Mullvad and used for customer role assumption on October 25. Shared VPN infrastructure, not conclusive alone Source: Wiz | Reported |
| SHA-256 Hashes | ||
65f061281c9a712e9502793dc4c5035cae490b84da05f7d4223cefa65ad396e4 |
sso_agent.iso, a bootable Linux image that collects system information and polls TCP port 23458 for shell scripts. Wiz observed no deployment
Source: Wiz
|
Reported |
| Cloud Identifiers | ||
805246085872 |
AWS account used by Qubole’s platform and the source account for every observed malicious role assumption. Activity outside AWS address ranges warrants review Source: Wiz | Reported |
AIDA3W7DIFLYOLTK63SFS |
IAM principal ID observed only in activity attributed to the threat actor Source: Wiz | Reported |
AIDAJYRRH5Y3A7AAJM2HW |
IAM principal ID used by the threat actor from October 18 and also used by legitimate Qubole platform traffic. Correlate with source IP and behavior Source: Wiz | Reported |
quiam-* |
Role session name pattern used by the threat actor, including quiam-enum and quiam-scope-full
Source: Wiz
|
Reported |
MITRE ATT&CK Mapping
| Tactic | Technique | Procedure | Status |
|---|---|---|---|
| Qubole Cross-Account Role Campaign | |||
| Resource Development | T1583.003 Acquire Infrastructure: Virtual Private Server | The threat actor worked primarily from an OVH virtual private server after beginning with a Mullvad VPN exit Source: Wiz, Sentra | Reported |
| Initial Access | T1078.004 Valid Accounts: Cloud Accounts | The threat actor used two IAM users from Qubole’s AWS account to obtain temporary credentials in customer environments Source: Wiz | Reported |
| Initial Access | T1199 Trusted Relationship | The threat actor assumed cross-account roles customers had created to trust Qubole’s AWS account. Every attributed request carried the correct external identifier and succeeded Source: Wiz, Sentra | Reported |
| Discovery | T1526 Cloud Service Discovery | The threat actor enumerated available AWS services, including Lambda, Secrets Manager, Systems Manager Parameter Store, and database services Source: Wiz, Sentra | Reported |
| Discovery | T1580 Cloud Infrastructure Discovery | The threat actor enumerated storage, compute instances, security groups, virtual private clouds, databases, database snapshots, and KMS keys Source: Wiz | Reported |
| Discovery | T1087.004 Account Discovery: Cloud Account |
The threat actor used AWS identity APIs including ListRoles while mapping accessible customer environments
Source: Wiz
|
Reported |
| Discovery | T1619 Cloud Storage Object Discovery |
The threat actor called ListBuckets, ListObjects, and HeadObject before selecting S3 data for collection
Source: Wiz
|
Reported |
| Credential Access | T1555.006 Credentials from Password Stores: Cloud Secrets Management Stores | The threat actor retrieved passwords and keys from AWS Secrets Manager and Systems Manager Parameter Store Source: Wiz, Sentra | Reported |
| Credential Access | T1552 Unsecured Credentials | The threat actor read Lambda environment variables and scraped EC2 user data for exposed credentials Source: Wiz, Sentra | Reported |
| Collection | T1530 Data from Cloud Storage |
The threat actor called GetObject to collect S3 data, in some cases downloading entire buckets, and later pulled production database backups
Source: Wiz, Sentra
|
Reported |
| n0n Expedia Extortion Claim | |||
| Collection | T1530 Data from Cloud Storage | n0n claims Expedia data from multiple cloud accounts, including loyalty, traveler profile, reservation, and workforce planning records. A published extract holds hundreds of thousands of email addresses tied to travel data. The source and full volume remain unverified Source: CyPro, Cyber Daily | Claimed |
| Impact | T1657 Financial Theft | n0n listed Expedia anonymously as Company #122 on November 3, named it on November 9 as “not paid, exposed,” and then published an extract of the claimed data Source: Ransomware.live, Cyber Daily | Reported |
Entities
| Entity | Role | Status |
|---|---|---|
| Threat Actors | ||
| Unattributed Qubole Cross-Account Threat ActorThreat actor, unidentified | Used IAM credentials from Qubole’s AWS account to assume customer roles and steal secrets and S3 data from October 14 to 26 Source: Wiz, Sentra | Reported |
| n0nData extortion threat actor, eCrime | Claimed responsibility for the Qubole intrusion and claimed Expedia data on its dedicated leak site. Its origin remains unknown despite unsupported media descriptions of a Russian group Source: Risky Business, CYFIRMA | Claimed |
| Unattributed Qubole Key TesterThreat actor, unidentified | Tested customer-owned AWS keys created for Qubole integrations from October 8 to 11 and compromised a small number of environments. Wiz has not linked this activity to the later campaign Source: Wiz | Reported |
| Organizations and Victims | ||
| Expedia GroupVictim | Confirmed threat actors used a compromised third-party data platform provider to reach Expedia systems in a supply chain attack affecting multiple companies. Expedia has not named the provider or confirmed n0n’s data claims Source: Skift | Confirmed |
| QuboleCloud data platform vendor, Idera subsidiary | Its AWS account was the source of the malicious role assumptions. The public evidence establishes compromised IAM credentials, not compromise of the entire Qubole account. No Qubole or Idera statement was found as of November 16 Source: Wiz, Sentra | Reported |
| Qubole CustomersVictims, unnamed | More than a dozen organizations were affected. The victims included former customers whose dormant Qubole integration roles remained active Source: SecurityWeek, Sentra | Reported |
| LiveRampClaimed victim | n0n claims 2.2 TB of identity graph feeds and 19.2 million hashed email identities. LiveRamp has not confirmed the claim, and no public reporting ties it to Qubole Source: CyPro | Claimed |
| DraftKingsClaimed victim | n0n claims 2020 advertising audience files with 7.4 million hashed email addresses. DraftKings has not confirmed the claim Source: Data Breach Class Action | Claimed |
| VF CorporationClaimed victim | n0n claims shopper chat transcripts and brand AI search caches. VF Corporation has not confirmed the claim, and no public reporting ties it to Qubole Source: DeXpose | Claimed |
| Tools and Infrastructure | ||
| TruffleHogOpen-source secrets scanner | Used in the Qubole campaign to validate harvested AWS keys, Microsoft Entra ID client secrets, and Okta client credentials Source: Wiz, Sentra | Reported |
sso_agent.isoBootable Linux image |
Hosted on the primary Qubole campaign server. It collects system information, polls for shell scripts, and reboots after 90 minutes. Wiz observed no deployment Source: Wiz | Reported |
| opencodeOpen-source AI coding agent | Observed in a separate September n0n workspace with 154 model entries across seven providers and more than US$7,216 in gateway spending. It has not been tied to Expedia or the Qubole campaign Source: SOCRadar | Reported |
| OVHHosting provider | Hosted the primary server used for role assumption, discovery, credential harvesting, and the undeployed ISO image Source: Wiz | Reported |
| MullvadVPN provider | Its shared exit nodes were used for the first observed role assumption and later activity. The service itself is not implicated Source: Wiz | Reported |
Related Activity
The activity below involved travel data collection, vendor compromise, or earlier n0n extortion. None was observed in the Qubole campaign or the Expedia breach.
| Item | Detail | Status |
|---|---|---|
| Marriott Starwood BreachIntrusion, disclosed 2018 | People familiar with Marriott’s investigation told Reuters Chinese hackers were suspected. The New York Times reported a suspected link to China’s Ministry of State Security, and Secretary of State Mike Pompeo appeared to confirm China’s role on television Source: GovInfoSecurity | Reported |
| REMIX KITTEN (aka APT39)Threat actor, nation state | The US Treasury sanctioned Rana Intelligence Computing in 2020 and said it worked for Iran’s Ministry of Intelligence and Security. Rana targeted or compromised about 15 US companies, primarily in the travel sector, to support tracking of people the ministry considered threats Source: CrowdStrike, US Treasury | Reported |
| Amadeus BreachIntrusion, 2026 | A three-year intrusion exposed airline booking data. Norway’s security authority said data from the incident had leaked Source: jark.me | Reported |
| CyberLogitec Maritime Listingsn0n extortion wave, October 2026 | n0n listed 38 port, shipping, and logistics organizations within 23 minutes. Cydome tied at least 21 to customers of CyberLogitec and assessed one vendor archive may have been divided into separate victim listings Source: CYFIRMA, Hellenic Shipping News | Reported |
| n0n September ActivityAI-assisted eCrime activity | SOCRadar recovered an exposed workspace showing a human-directed effort supported by opencode. The documented compromises relied on weak or stolen credentials, and SOCRadar observed no fully autonomous attack system Source: SOCRadar | Reported |
| Cursor and Fireworks AISupply chain intrusion, 2026 | A stolen Fireworks AI vendor key led to trojanized Claude Code extensions on Cursor’s marketplace Source: BleepingComputer | Reported |
Intelligence Gaps
| Gap | Why It Matters | What Would Close It |
|---|---|---|
| Identity of Expedia’s third-party provider | Qubole is the strongest candidate, but Expedia has not named it and no researcher has placed Expedia among the Qubole victims. SOCRadar also notes Expedia-linked credentials in infostealer logs before the claim | An Expedia disclosure or regulatory filing, or confirmation from Wiz, Sentra, Mandiant, Qubole, or Idera |
| Attribution of the Qubole campaign to n0n | n0n has claimed responsibility, but Wiz identified no links to previously reported activity and has not independently attributed the campaign | Infrastructure, credential, tooling, or data evidence connecting n0n to the October 14 to 26 sessions |
| How the Qubole AWS credentials were obtained | The original credential compromise sits upstream of every customer role assumption and may expose additional customers | A Qubole or Idera disclosure, or forensic reporting identifying the credential source |
| Relationship between the October 8 and October 14 activity | Two threat actors held different forms of Qubole-related access in the same month, raising the possibility of a shared source or credential market | Infrastructure, tooling, timing, or credential provenance connecting the two activity sets |
| Authenticity and scale of the Expedia data | A published extract covers a fraction of the claim, and Expedia has not verified it. The row counts do not measure unique people and may combine transactions, snapshots, and duplicate identities | Verified samples, Expedia notifications, or regulator filings describing the exposed data and affected population |
| Object-level logging | Missing S3 data events and server access logs in several environments prevent responders from proving or excluding the claimed download volume | S3 data events, server access logs, or KMS decrypt records from the affected environments |
| Status of the claimed Expedia credentials | Valid and sufficiently privileged credentials for data, source code, or production cloud environments could enable re-entry or code tampering | Expedia confirmation of credential rotation, privilege scope, and any observed use |
| Ownership of the 35 claimed loyalty programs | Partner loyalty members could be affected without knowing Expedia processed their bookings | Expedia or partner notifications identifying the programs and affected datasets |
Intended use of sso_agent.iso |
The image supports remote script execution and appears designed for Windows systems, but Wiz observed no deployment and cannot connect it to a specific objective | Deployment telemetry, command payloads, or victim evidence showing how the image was used |
| Basis for the Russian attribution | A nationality label shapes policy and public response, while the published language and infrastructure evidence does not identify an origin | Infrastructure ownership, payment records, threat actor communications, or other evidence connecting n0n to a location |
| Other Qubole-linked listings | LiveRamp, DraftKings, and VF Corporation hold data types associated with analytical platforms, but none has been tied publicly to Qubole | Victim confirmation or researcher matching of claimed samples to Qubole-accessible environments |
| Fate of authentic Expedia data | n0n has published an extract. Publication or private sale of the rest would decide who can acquire the full dataset, including criminal or foreign intelligence buyers | DLS monitoring, torrent tracking, underground market sightings, or confirmed acquisition by another party |