← Breach analysis by the Dual Vulnerabilities framework

Odido customer data theft (ShinyHunters, 2026)

2026 · Netherlands · J - Information and communication · Customer data theft via vishing and bulk CRM export

Analysed 2026-11-07 by Hermes

Open this case in the tool →

What happened

Customer data theft via vishing and bulk CRM export

Dual vulnerability type

Expertise to Execution

Evidence coverage: 3/6 stages, weight 24/60

Decisive stage: egress (clearest evidence) · Alternative: Strategy to Execution

Why this classification

(1) Odido, the telecom holding the personal data of its ~8 million customers in a Salesforce CRM, sat inside a sector-wide expert warning loop: Salesforce had flagged this identity-compromise method in an October 2025 security update and again in late January 2026, explicitly noting the type of two-step verification used could be bypassed, and the FBI and Mandiant had published warnings on the same ShinyHunters campaign; internally, staff were trained and two-step verification was deployed. (2) In practice, on 5 and 6 February, vishing calls to the helpdesk pretending to be IT staff yielded an employee's credentials and defeated the extra verification step; the intruder was detected and blocked within an hour, but that single helpdesk account could bulk-download the data of about 6.2 million accounts, no alarm fired during the download, the same-day forensic review (with an external security firm) concluded nothing had been stolen, and Odido only learned of the theft from the attackers' ransom message two days later - the leak began on the dark web on 26 February and the full scope (business customers, sensitive service notes) emerged only after the complete dump on 1 March. (3) The classification rests on the egress stage: the knowledge that should have prevented or revealed the bulk export sat entirely at the material level - vendor security guidance, FBI and Mandiant advisories naming this exact method - yet it never materialised in operations: no egress alert, no export volume detection, and a post-intrusion forensic sweep that missed an hour-long mass download. The chain broke in execution, not at the decision level: no source shows leadership deferring or accepting the risk (NOS: whether Odido acted on the warnings is unknown), so this is a within-material-column Expertise to Execution gap; the entry gap (bypassable verification type despite explicit vendor warning) and the containment gap (one helpdesk credential spanning the whole customer base) point the same way, so no tie-break was needed. A secondary Strategy to Execution pattern sits behind this: the AP is examining whether retention of ID numbers, bank account numbers and sensitive notes on vulnerable customers exceeded what is allowed, which would give the over-broad export a doctrinal origin, but there is no finding yet and the condition is therefore not coded. (4) RM22 (Expertise to Execution lens): the organisation kept an authentication stack whose extra verification step was of the very type Salesforce had warned can be bypassed, so identity controls did not hold against the warned campaign (causal role: Enabling). RM24 (Expertise to Execution lens): live Dutch-language vishing against the helpdesk succeeded in stealing credentials on a fake copy of the work environment, showing training and verification routines did not absorb the technique the sector had been warned about (causal role: Enabling). RM41 (Expertise to Execution lens): after the login the account's privileges spanned the entire customer database, so blast radius equalled one compromised helpdesk session rather than that employee's task scope (causal role: Amplifying). RM35 (Expertise to Execution lens): a ~6.2-million-account bulk API export left the Salesforce environment without any egress alert or interruption in an environment where the vendor and agencies had pointed to exactly this monitoring requirement (causal role: Amplifying).

Evidence by stage

Risk markers

Sources

Notes

cyberdefensemagazine row, Human Verified; description matches the stated victim (Odido, ~6 million accounts, attacker leak). Timeline: Odido's own disclosure dates the vishing attacks 5 and 6 February 2026; some third-party write-ups place a first call on 3 February - the victim's dates are used. The claim that 'Odido does not apply the principle of least privilege' comes via ShinyHunters as reported by an expert blog; the containment gap is instead grounded in NOS's own sourcing (too broadly configured employee access) and the CEO's account of the download window. Odido denies that account passwords leaked; customer-service voice-verification 'code words' were acknowledged to be in the leak and that verification method was discontinued. Two regulator investigations (security of the customer system; lawfulness of retention) were ongoing with no findings as of NOS's May 2026 interview, so RM16/RM39 are not coded; the CEO acknowledged follow-up communication was too sparse but the theft was self-disclosed on 12 February before any leak. NOS was denied insight into the internal forensic report, so the detection-stage reading rests on the CEO's public account. Unattended run: nothing requiring consent was performed; only public pages were read. Link check: reuters.com blocks automated access (HTTP 401); the page was opened and read during research.