← Breach analysis by the Dual Vulnerabilities framework

Operation Skeleton Key IP theft (Chimera, 2018)

2018 · Taiwan · C - Manufacturing · Industrial-espionage data theft via long-dwell APT intrusions

Analysed 2026-11-06 by Hermes

Open this case in the tool →

What happened

Industrial-espionage data theft via long-dwell APT intrusions

Dual vulnerability type

Expertise to Execution

Evidence coverage: 4/6 stages, weight 28/60

Decisive stage: detection (clearest evidence) · Alternative: none defensible

Why this classification

(1) Semiconductor vendors holding crown-jewel IP can be expected to protect remote access, the directory, monitoring and egress with controls their own security specialists know - and the sources show these remedies were within reach for this victim group: Case A, which had subscribed to a continuous threat-hunting service, had its intrusion identified within days. (2) In practice, CyCraft's forensic report shows an attacker logging into a VPN with a valid account whose password was likely bought or harvested from a separate breach, moving laterally with valid accounts and scheduled tasks, archiving ntds.dit and the registry for offline cracking, injecting a skeleton key into the domain controllers to gain domain-wide log-in capability, and staging chip documents, SDKs and source code with a disguised RAR tool - all in Case B between 07-10-2018 and 18-11-2019, until the intrusion was found by chance during a network upgrade; the initial access point could no longer be pinpointed. CyCraft explicitly documents an absence of AD monitoring systems among the vendors it investigated. (3) The classification rests on the detection stage: the expectation that was not met sits with technical specialists (monitoring, detection, egress control - capabilities demonstrably available, as the threat-hunting subscription shows), and the chain broke in execution, because the capability was simply never realised in operations; nothing in the sources shows a strategic artefact, a leadership decision, a deferral or an ignored expert warning, so no Strategy or Expertise to Leadership variant is evidenced. All other documented gaps (entry, containment, egress) point to the same material-column gap, so the case is a coherent Expertise to Execution, not merely weak security: each gap is a known, available remedy left unexecuted. The entry vector in Case B is unknown and not load-bearing. RM22 (Expertise to Execution lens): the VPN login succeeded with a valid account whose password was believed acquired from a separate data breach, with nothing in the forensic record showing secondary authentication or leaked-credential monitoring - identity controls known to specialists were not implemented (causal role: Enabling). RM41 (Expertise to Execution lens): after initial access the attacker copied ntds.dit for offline cracking and injected a skeleton key into the domain controllers, turning one intrusion into domain-wide log-in capability - directory hygiene and privilege tiering known to experts were left unexecuted (causal role: Amplifying). RM34 (Expertise to Execution lens): CyCraft documents that the vendors lacked AD monitoring systems and white-listing was bypassed with living-off-the-land binaries, and Case B's intrusion ran over 13 months undetected, found only by chance during an upgrade - available detection capability was never built or tuned (causal role: Obscuring). RM35 (Expertise to Execution lens): archives of chip documents, SDKs and source code were repeatedly staged with a disguised RAR binary and exfiltrated across more than a year with no victim-side detection - egress monitoring and prevention known to specialists were not operationalised (causal role: Amplifying).

Evidence by stage

Risk markers

Sources

Notes

Case file carries no receiver_name; 'Operation Skeleton Key' is the EuRepoC incident name used as the victim label per the task - the actual victims are at least seven Taiwanese semiconductor companies, unnamed by CyCraft (Wired, Black Hat 2020 coverage). The cycraft.com PDF in the case file returns 404 ('Not Found'); the report was read via the cited Internet Archive snapshot of the CyCraft whitepaper (opened during this analysis). TSMC's August 2018 worm outbreak is cited only as background in the CyCraft report and was characterised by TSMC itself as a failure to follow installation protocol, not as a targeted state attack; no source ties it to this row's victim set, so it was not coded here. Attribution to Chimera/Winnti comes from the IT-security community (CyCraft states the Winnti link is likely but not certain). Stage origin/break classifications are read from the vendor forensic report; victim-side policy or governance evidence is absent from all public sources because the victims are unnamed. Country from eurepoc_extra.receiver_country; Year 2018 from eurepoc_extra.start_date (campaign documented 2018-2019, disclosed 2020).