← Breach analysis by the Dual Vulnerabilities framework

New York Taxis FOIL Trip-Data De-anonymisation (None, 2014)

2014 · United States · O - Public administration and defence · Personal data exposure in an improperly anonymised public data release

Analysed 2026-11-06 by Hermes

Open this case in the tool →

What happened

Personal data exposure in an improperly anonymised public data release

Dual vulnerability type

Strategy to Execution

Evidence coverage: 1/6 stages, weight 10/60

Decisive stage: egress (clearest evidence) · Alternative: Expertise to Execution

Why this classification

(1) Strategic expectation: when the New York City Taxi and Limousine Commission handed over the 2013 trip and fare records covering 173 million rides under a FOIL request, the release decision itself carried the requirement to protect personal privacy - the agency deliberately pseudonymised the two identifying fields, replacing medallion and hack licence numbers with one-way MD5 hashes, and its later statement defended anonymisation of the process as the safeguard. (2) In practice: the pseudonymisation was architecturally ineffective. Medallion and licence numbers follow rigid formats, leaving only about 24 million possible values; an external developer hashed that whole space in minutes and within about two hours had tied every one of the 173 million trips to a named driver and a specific taxi, with drivers' incomes and inferred home locations following; the published file stayed re-identifiable for months, with even individual passenger rides reconstructed later. (3) This is a Strategy to Execution gap: the expectation sat at the strategic level (the agency's own decided intent to anonymise before release, a decision it implemented in the release process itself), and the chain broke at execution - the chosen operational scheme simply did not achieve the protection that had been decided on, while standard alternatives (keyed hash with a secret key, encryption, random surrogate identifiers) would have. There is no entry vector at all in this case - no intrusion, no attacker - so the classification rests wholly on the egress/data-governance stage. RM16 (Strategy to Execution lens): the privacy obligation attached to the release was never implemented in the publication process - identifiers were published in a form any laptop owner could reverse, so far more sensitive data was exposed than the release required (causal role: Amplifying). RM26 (Strategy to Execution lens): the data-protection requirement was implemented with an inadequate technique - unkeyed MD5 pseudonymisation over a tiny, structured identifier space instead of encryption or a keyed hash, exactly the weakness NIST lists as avoidable (causal role: Amplifying). RM46 (Strategy to Execution lens): the safeguards for publishing a 20 GB personal dataset - verification of the anonymisation, four-eyes or legal review - were not implemented: outsiders found the flaw, and TLC would not say whether legal had reviewed the release (causal role: Enabling).

Evidence by stage

Risk markers

Sources

Notes

Victim recorded as written in the case file; the organisation that released the data is the New York City Taxi and Limousine Commission (a city regulator, hence sector O), and the exposed population is taxi drivers (passengers secondarily). Not an intrusion case: the agency itself published the data under FOIL, so the case is classified from the egress/data-governance stage under the whole-lifecycle rule; entry, containment and recovery are not in play. The case file's source_url is a malformed Medium short link (a stray ' /' appended); the canonical article (Pandurangan, 'On Taxis and Rainbows', 21-06-2014) was opened instead and is cited. The dataset's 173 million trips are the calendar year 2013; Year is set to 2014, the year of the release and the de-anonymisation. An earlier 2014 FOIL release of the same 2013 data to the New York Times was reportedly also de-anonymised, but the NYT article is not accessible and no opened source confirms it, so nothing is claimed from it and RM21 (recurring weakness) is not coded. Detection and response stages were left as no documented gap rather than inferred. Unattended run: no action required consent; no leak sites or leaked data were used - the released dataset is public FOIL open data and only press, the releaser's account and the NIST report were read.