← Breach analysis by the Dual Vulnerabilities framework

GitHub DDoS attack (suspected Chinese state actor, 2015)

2015 · United States · J - Information and communication · Distributed denial-of-service (DDoS) attack

Analysed 2026-11-07 by Hermes

Open this case in the tool →

What happened

Distributed denial-of-service (DDoS) attack

Dual vulnerability type

Expertise to Execution

Evidence coverage: 1/6 stages, weight 8/60

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

Why this classification

(1) GitHub's business is always-on availability of a core code-infrastructure platform, and the expectation that applies to that - defending service capacity against volumetric flood - is specialist (material) knowledge: GitHub's CEO stated the site is attacked once a month or more and had monitoring, automated systems and an around-the-clock network ops team. (2) In practice, from 26 March 2015 the Great Cannon operation hijacked third-party Baidu JavaScript in the Chinese backbone to enlist millions of bystander browsers, flooding the greatfire and cn-nytimes repositories and then Gist, Pages and assets through 18 to 20 evolving attack profiles over five days; GitHub's in-house mitigation deflected most attack traffic but users saw intermittent outages platform-wide until roughly hour 113-118. (3) The classification rests on the recovery/resilience stage; the entry vector is explicitly not load-bearing (a state man-on-the-side injection in a foreign backbone, no victim-side remedy), detection was same-day and response/disclosure were continuous and self-initiated, so the one documented victim-side gap is that GitHub's own-capacity defence architecture, though decided, staffed and executing, could not keep the service fully available against nation-scale traffic: expertise existed and was applied, but capacity and architecture constraints prevented its full operationalisation - a material-origin, execution-break gap (Expertise to Execution). No source shows leadership deferring or underfunding availability protection, which is why this is not a leadership break; no strategic artefact is documented that operations failed to implement, which is why it is not Strategy to Execution. RM32 (Expertise to Execution lens): GitHub's experts could not scale or protect the service under the flood because their defence depended on their own network architecture and capacity rather than supplier-side scrubbing at the scale required - traffic volumes of roughly 700,000 requests per second and a five-day outage timeline are documented (causal role: Enabling). RM04 (Expertise to Execution lens): all of GitHub's critical services (github.com, Gist, Pages, assets) hung off the same front-door infrastructure and IP block, so an attack aimed at two repositories degraded the whole platform for days, and the experts could not decouple that dependency under the platform constraints of the service (causal role: Amplifying).

Evidence by stage

Risk markers

Sources

Notes

Dataset date caveat: eurepoc_extra gives start 19.03.2015 / end 23.03.2015, but press and GitHub's own status page date the github.com flood from 26 March to about 1 April 2015; the 19-23 March window matches the earlier attack on GreatFire.org's own servers, which preceded and morphed into the GitHub attack - the described incident (GitHub DDoS) is dated 2015 from the victim's own pages. Attribution caveat: the case file itself lists initiator Unknown / anonymous media statements / IT-security-community attribution; Baidu and the Chinese government denied involvement; the Citizen Lab / USENIX forensic reports attribute the attack tool (Great Cannon) to Chinese authorities technically but there is no official government attribution - the Actor is therefore stated as suspected. The eWeek availability-partner remark is expert sector commentary about DDoS defence in general quoted in this incident's coverage, not a GitHub-specific claim; the origin/break reading is therefore recorded as inference. Description matches the stated victim (GitHub); no data-quality mismatch. No related_rows in the case file.