← Breach analysis by the Dual Vulnerabilities framework

Okta Support System Breach (Unattributed, 2023)

2023 · United States · J - Information and communication · Breach of support case management system, data theft and customer session hijacking

Analysed 2026-11-07 by Hermes

Open this case in the tool →

What happened

Breach of support case management system, data theft and customer session hijacking

Dual vulnerability type

Expertise to Execution

Evidence coverage: 3/6 stages, weight 22/60

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

Why this classification

(1) Okta is an identity-security vendor whose own products enforce credentials and sessions; the expectation that applied to it internally was standard operational execution of those controls - protected privileged accounts, monitored support-system activity - and its own RCA shows a resourced security operations team actively investigating customer reports from 29 September 2023. (2) In practice, between 28 September and 17 October 2023 a threat actor logged into Okta's customer support case management system with the password of a service account whose credentials were stored in that system and had been saved into an employee's personal Google account; Okta says compromise of the employee's personal Google account or device is the most likely exposure route. Okta did not notice: for 14 days, while actively investigating, it did not identify the suspicious downloads in our logs because its analysts examined only case-linked log events, not the different events the attacker's Files-tab access produced; detection came only from the attacker IP BeyondTrust handed over on 13 October, 11 days after BeyondTrust first reported the hijack attempt and after repeated escalations. The actor downloaded files of 134 customers (under 1%), used session tokens from HAR files to hijack 5 customers' sessions, and had also exported an unfiltered report with the names and email addresses of all support system users - found only in a November review and disclosed on 29 November. Production was unaffected and nothing was destroyed, so recovery was not in play, and disclosure itself was prompt and self-initiated once facts were established. (3) The classification rests on the detection stage, the best-documented origin-break pair: the knowledge that went unapplied was technical-operational (which vendor log event types capture file access, how to tune support-system monitoring) and the team holding it was engaged and resourced, yet operations did not realise it - origin material, break execution, Expertise to Execution; the entry gap sits in the same cell (known service-account credential controls not executed internally), while no source shows a doctrinal expectation leadership deferred (against Strategy to Leadership/Execution) or an expert warning leaders overrode (against Expertise to Leadership). The entry mechanism is documented but its exposure route stays the most likely avenue rather than confirmed, and is not load-bearing for the type. RM22 (Expertise to Execution lens): identity specialists at an identity vendor did not have credential lifecycle controls over their own privileged service account - its password lived in the support system and synced into an employee's personal Google profile, letting the attacker log in with it, a known remedy left unexecuted (causal role: Enabling). RM34 (Expertise to Execution lens): the monitoring available on the support system was not configured or tuned to detect the activity - Okta's RCA states that for a period of 14 days, while actively investigating, Okta did not identify suspicious downloads in our logs because only the wrong class of log events was examined (causal role: Obscuring). RM35 (Expertise to Execution lens): egress monitoring was not tuned to interrupt bulk export inside the support system - the unfiltered report of every support user's name and email, plus customer files, left on 28 September unnoticed until the November review, which widened the disclosed impact from under 1% of customers to all of them (causal role: Amplifying).

Evidence by stage

Risk markers

Sources

Notes

Source is Okta's own RCA, customer disclosure and independent forensics; the incident is well documented on the victim side. Data quality: the row description conflates the figures - per Okta's RCA 134 customers (under 1% of its roughly 18,400 business customers) had support files accessed and 5 customers had sessions hijacked, while the names and email addresses of all support-system users came from a separate unfiltered report export disclosed only on 29 November 2023; the 5 hijackings match Okta's figure. The case file's source_url carries a trailing space (https://sec.okta.com/harfiles /); the live page now shows Okta's investigation-closure post (Stroz Friedberg found no further malicious activity; Okta changed the support system's data retention policy). Disclosure: the initial advisory (20 Oct) and RCA (3 Nov) reported the sub-1% figure; the scope widened to all support users on 29 Nov after a review found the missed unfiltered report - Okta disclosed promptly on discovery, so RM39 is not coded and the widening is recorded at the egress stage. Attribution is unconfirmed (third-party assessments suggested possible Scattered Spider involvement); it is not a DV claim and does not affect confidence. Related row 879 (Identity and access management provider Okta): its case file is not present in the run's cases folder and no result-879.json exists in the output directory, so no comparison of that row was possible. Link check: cybersecuritydive.com blocks automated access (HTTP 403); the page was opened and read during research. Link check: beyondtrust.com blocks automated access (HTTP 403); the page was opened and read during research.