← Breach analysis by the Dual Vulnerabilities framework

Carnegie Mellon University data breach (Unattributed, 2023)

2023 · United States · P - Education · Data theft via unauthorized access to a computer system

Analysed 2026-11-06 by Hermes

Open this case in the tool →

What happened

Data theft via unauthorized access to a computer system

Dual vulnerability type

No DV evidenced

Alternative: Strategy to Leadership

Why this classification

(1) Strategic expectation: Carnegie Mellon University is a private research university holding identity data of its students, employees, applicants and contractors, and it runs a dedicated Information Security Office whose documented remit covers detection, investigation and lawful breach notification - the expectation it was held to is exactly the one its notification record describes: notice to affected individuals with what happened, what was involved and what was done. (2) In practice: in August 2023 an unauthorized third party briefly accessed one university computer system holding files with names, social security numbers and dates of birth of 7,343 current and former students, employees, applicants and contractors. The Information Security Office detected the suspicious activity on 25 August 2023 itself, secured the system within hours and engaged law enforcement immediately; the file-by-file scoping with law enforcement concluded on 4 December 2023, notification letters went out on 12 January 2024 with a 24-month Experian IdentityWorks offer, and CMU found no evidence of fraud or inappropriate use. The entry vector was never disclosed and the actor was never publicly attributed; an April 2025 class action alleges inadequate security procedures and delayed notice, but those are pleaded allegations and no court or regulator finding of failing appears in the opened record. (3) Why this is a confirmed negative rather than a DV: control performance is positively documented at three lifecycle stages - detection (the victim's own monitoring surfaced the intrusion), containment (secured within hours; access to the copied files disabled; blast radius limited to specific files on one system) and response (law-enforcement-coordinated investigation completed, self-initiated direct notification of every identified affected person, credit monitoring offered) - and no opened source documents a victim-side gap at any stage, so no origin-and-break pair can be answered for any stage; the undetermined entry route alone does not create a gap. Conditions the evidence shows are recorded as risk markers only: after months of forensic review CMU still could not determine whether the files were taken - the final notice says only that the intruder may have copied them - and the possible copying of sensitive files was reconstructed after the fact rather than detected or interrupted while it happened. Two further conditions are described rather than coded: identity data for 7,343 people resided on the compromised system in files an intruder with system access could copy quickly (an RM16-type condition, but no source shows data-governance obligations were not operationalised), and roughly 4.5 months passed from detection to notification (an RM39-type timeline, but the letter documents the reasons - law-enforcement-assisted file identification and contact confirmation - the disclosure was self-initiated and legally permissible, so the RM39 exclusion applies). Recovery was not in play: no disruption, no data corruption. The entry stage yields no assessable mechanism and is not load-bearing.

Risk markers

Sources

Notes

Source caveats: (a) the notification letter opened from mass.gov is the sample (L01) copy CMU filed with state attorneys general (Massachusetts breach 2024-070), a primary victim disclosure; the same text is on the California AG site; (b) the CBS Pittsburgh page reproduces CMU's full statement verbatim (secured within hours, law enforcement engaged, 7,343 notified) - the govtech/Tribune-Review article quotes the same statement; (c) the entry vector and actor are undisclosed in every opened source, so entry yields no assessable mechanism; (d) the Dermenjian complaint's claims of inadequate security and a 6-month notification delay are plaintiff allegations pleaded 'on information and belief'; the CourtListener docket record opened for 2:25-cv-00459 (metadata last modified 28 October 2025) and the April 2025 scheduling order visible there show no ruling on those allegations as of this analysis; (e) RM33 and RM35 are coded at inference level from the victim's own letter: the final notice could not state whether files were copied (only may have) and the possible exfiltration surfaced through after-the-fact review, not an interrupted transfer; no source describes the logging or DLP state, and neither condition is claimed as a DV gap; (f) the Pfingsten v. CMU settlement visible in searches is a Spring-2020 tuition case, unrelated to this breach. Data quality: the Victim field repeats the name ('Carnegie Mellon University;Carnegie Mellon University'); eurepoc_extra start_date 01.08.2023 and legal_response_date 2023-8-01 are not supported by any opened source, which dates detection to 25.08.2023 - Year 2023 is unaffected. The description matches the stated victim. A breach-database site (csidb.net) appeared in searches and was not used, per the research boundaries; no leak site or dark-web source was consulted. Link check: mass.gov blocks automated access (HTTP 403); the page was opened and read during research. All four cited URLs were opened and read during this task (PDFs via download + pdftotext, articles via direct fetch).