← Breach analysis by the Dual Vulnerabilities framework

Cock.li Webmail Database Theft (Unattributed, 2025)

2025 · Germany · J - Information and communication · Data theft via webmail software vulnerability

Analysed 2026-11-07 by Hermes

Open this case in the tool →

What happened

Data theft via webmail software vulnerability

Dual vulnerability type

Expertise to Execution

Evidence coverage: 3/6 stages, weight 14/60

Decisive stage: entry (clearest evidence) · Alternative: Strategy to Leadership

Why this classification

(1) Cock.li is an established (since 2013) German free e-mail hosting provider with ~1.4M users whose standing rests on a privacy-focused, technically careful image - its own statement calls its past choices 'an otherwise cautious security philosophy and structure'; for such a provider the baseline expectation is that internet-facing webmail software is kept patched and eventually replaced or isolated. (2) In practice its Roundcube webmail kept running aged code and was exploited - the victim's own best assessment is CVE-2021-44026, a SQL injection remediated in 2021 - and the entire Roundcube 'users' and 'contacts' tables (~1,023,800 users who had logged in since 2016, ~93,000 contacts) were copied out without ever being detected; the leak surfaced only when the attacker offered the tables for sale for 1 BTC in June 2025. Passwords, e-mail content and IP addresses were not in the stolen tables, so the blast radius stayed within what the webmail front end could reach. (3) This is a knowing-doing gap inside the material column: the remedy was public technical knowledge for three and a half years (NVD publication Nov 2021, vendor patch, CISA KEV listing June 2023), and the break is in execution, not strategy - the victim's own disclosure concedes faster updating would probably have prevented the leak and that Roundcube should not have been running at all, while its site log documents a 3-5-person unpaid volunteer team that had 'slowly taken steps back' from maintaining the service. Expertise and the remedy existed but capacity and legacy kept them from being realised; no strategic artefact that was ignored is documented, and no expert-to-leadership warning chain exists in a volunteer-run service. The exact entry vector is the victim's stated guess and is not the load-bearing point; what is documented is the non-execution of a known remedy, plus absent detection and egress monitoring (contributing). The classification rests on the entry stage (clearest evidence; the detection and egress stages rest on inference and cannot win a tie-break, and they point to the same DV-Type anyway). (4) RM17 (Expertise to Execution lens): the ageing Roundcube installation stayed in exposed service for years because the volunteer operators kept it by choice instead of replacing or isolating it ('our choice in software has reflected the fact that e-mail has been mostly unchanged for over 40 years'), and it was retired only wholesale in June 2025 (causal role: Enabling). RM18 (Expertise to Execution lens): the suspected SQL injection was known and remediable from November 2021, but update execution lagged - the victim itself: 'Could we have prevented this leak by updating Roundcube faster? Probably!' (causal role: Enabling). RM34 (Expertise to Execution lens): monitoring and alert handling were never configured or staffed, so the intrusion went undetected and only the attacker's public sale revealed it, leaving dwell time unknown and possibly years (causal role: Obscuring). RM08 (Expertise to Execution lens): the remedy was knowable but could not be operationalised - the site log documents 3-5 unpaid volunteers taking leave from paid jobs just to keep the service running, a documented capacity shortfall (causal role: Enabling).

Evidence by stage

Risk markers

Sources

Notes

informationisbeautiful row; the description matches the victim. The word 'retired' is retrospective: Roundcube was pulled from the service on 15-06-2025, just before the leak surfaced, and the breach was of the instance still running until then. Exploitation path is unconfirmed: the victim states CVE-2021-44026 as a guess and chose not to reconstruct the exact exfiltration means, and the data may have 'been held onto for a while' - hence primary-open for the entry and disclosure claims. There is tension in the record on whether the vulnerable version was still patched-current before the theft: the victim says the affected versions were 'updated long ago' yet attributes the leak to that CVE and admits faster updating would probably have prevented it; Tech Monitor reports the provider 'ceased using the vulnerable version long ago' but calls the platform 'obsolete'. The classification rests on the victim's own acknowledgement that update/replace execution lagged, not on a confirmed exploit date. The 2015-2016 seizures of Cock.li disks by German authorities were law-enforcement actions (school bomb-hoax investigation), not a security breach, so no earlier comparable incident and no RM21. No regulator finding, fine or GDPR notification to authorities is documented in the reporting, so RM03 was not coded; RM35 was not coded because the duty of egress visibility rests on inference rather than a source - the absent egress monitoring is described in the rationale; RM33/RM36/RM39 not coded: scope of leaked data was established and disclosure, once the leak was known, was prompt and accurate. The victim's statement is no longer on the live site (mail.cock.li now serves its replacement webmail app); it is cited via the exact Wayback snapshots opened during this analysis. Country per reporting ('operates out of Germany'); the team is pseudonymous ('Vincent Canfield') and unpaid. Actor unattributed: a threat actor advertised the tables on X and underground forums; attribution is not part of the DV assessment.